theluyuan/DiffSynth-Studio
1
0
Fork 0
mirror of https://github.com/modelscope/DiffSynth-Studio.git synced 2026-03-18 22:08:13 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'main' into firered

Browse Source
This commit is contained in:
Zhongjie Duan
2026-02-15 14:12:44 +08:00
committed by GitHub
parent b08bc1470d 94b57e9677
commit 863a6ba597
84 changed files with 2017 additions and 575 deletions
Download Patch File Download Diff File Expand all files Collapse all files

62
examples/ltx2/model_inference/LTX-2-T2AV-Camera-Control-Dolly-In.py Normal file
View File

@@ -0,0 +1,62 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.bfloat16,
"offload_device": "cpu",
"onload_dtype": torch.bfloat16,
"onload_device": "cuda",
"preparing_dtype": torch.bfloat16,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Dolly-In", origin_file_pattern="ltx-2-19b-lora-camera-control-dolly-in.safetensors"),
)
prompt = "Dolly-in shot: A cheerful girl smiles brightly and says, 'I enjoy working with Diffsynth-Studio, it's a perfect framework.' The camera smoothly moves closer to her face, highlighting her enthusiasm and sincerity."
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path='ltx2_dolly_in_lora.mp4',
fps=24,
audio_sample_rate=24000,
)

62
examples/ltx2/model_inference/LTX-2-T2AV-Camera-Control-Dolly-Left.py Normal file
View File

@@ -0,0 +1,62 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.bfloat16,
"offload_device": "cpu",
"onload_dtype": torch.bfloat16,
"onload_device": "cuda",
"preparing_dtype": torch.bfloat16,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Dolly-Left", origin_file_pattern="ltx-2-19b-lora-camera-control-dolly-left.safetensors"),
)
prompt = "Dolly-left shot: A joyful young woman sits at a minimalist desk with a laptop running Diffsynth-Studio, code and generative visuals glowing on screen. She turns slightly toward the camera and says with a smile, 'I enjoy working with Diffsynth-Studio, it's a perfect framework.' As she speaks, the camera smoothly dollies left, revealing a wall of framed open-source project posters, a whiteboard covered in neural network sketches, and a shelf stacked with AI/graphics books beside her."
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path='ltx2_dolly_left_lora.mp4',
fps=24,
audio_sample_rate=24000,
)

63
examples/ltx2/model_inference/LTX-2-T2AV-Camera-Control-Dolly-Out.py Normal file
View File

@@ -0,0 +1,63 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.bfloat16,
"offload_device": "cpu",
"onload_dtype": torch.bfloat16,
"onload_device": "cuda",
"preparing_dtype": torch.bfloat16,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Dolly-Out", origin_file_pattern="ltx-2-19b-lora-camera-control-dolly-out.safetensors"),
)
prompt = "Dolly-out shot: A joyful young woman smiles warmly and says: 'I enjoy working with Diffsynth-Studio, it's a perfect framework.' As she speaks, the camera slowly dollies out, revealing a bright, modern creative studio filled with plants, whiteboards full of diagrams, and soft natural light from large windows."
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path=f'ltx2_dolly_out.mp4',
fps=24,
audio_sample_rate=24000,
)

62
examples/ltx2/model_inference/LTX-2-T2AV-Camera-Control-Dolly-Right.py Normal file
View File

@@ -0,0 +1,62 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.bfloat16,
"offload_device": "cpu",
"onload_dtype": torch.bfloat16,
"onload_device": "cuda",
"preparing_dtype": torch.bfloat16,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Dolly-Right", origin_file_pattern="ltx-2-19b-lora-camera-control-dolly-right.safetensors"),
)
prompt = "Dolly-right shot: A happy girl looks up and says happily, 'I enjoy working with Diffsynth-Studio, it's a perfect framework.' She sits before a sunlit café table, her open laptop displaying the Github interface. The camera glides right to show a barista crafting coffee in the background, shelves of artisan beans, and a chalkboard menu softly blurred in the bokeh."
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path='ltx2_dolly_right_lora.mp4',
fps=24,
audio_sample_rate=24000,
)

72
examples/ltx2/model_inference/LTX-2-T2AV-Camera-Control-Jib-Down.py Normal file
View File

@@ -0,0 +1,72 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.bfloat16,
"offload_device": "cpu",
"onload_dtype": torch.bfloat16,
"onload_device": "cuda",
"preparing_dtype": torch.bfloat16,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Jib-Down", origin_file_pattern="ltx-2-19b-lora-camera-control-jib-down.safetensors"),
)
prompt = (
"A girl is very happy, standing on a clean studio floor with soft ambient lighting. "
"She is speaking directly to the camera: “I enjoy working with Diffsynth-Studio, it's a perfect framework.” "
"The shot begins with a medium close-up framing her from the waist up. As the camera performs a smooth jib-down movement—"
"descending vertically downward—it gradually reveals more of the lower portion of the scene. "
"During the descent, the following elements become visible near the bottom of the frame: "
"- The polished concrete floor with subtle reflections of the girl’s shoes, "
"- A small branded mat labeled “Diffsynth-Studio” placed just beneath her feet, "
"- The lower part of a sleek workstation desk with a glowing logo on its front panel, partially hidden at the start but fully revealed as the camera lowers. "
"This downward motion provides a dynamic reveal of contextual details that reinforce the professional and creative environment, "
"while maintaining focus on the girl’s enthusiastic expression throughout."
)
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path='ltx2_camera_jib_down.mp4',
fps=24,
audio_sample_rate=24000,
)

66
examples/ltx2/model_inference/LTX-2-T2AV-Camera-Control-Jib-Up.py Normal file
View File

@@ -0,0 +1,66 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.bfloat16,
"offload_device": "cpu",
"onload_dtype": torch.bfloat16,
"onload_device": "cuda",
"preparing_dtype": torch.bfloat16,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Jib-Up", origin_file_pattern="ltx-2-19b-lora-camera-control-jib-up.safetensors"),
)
prompt = (
"A girl stands happily at a sleek desk with a glowing 'Diffsynth-Studio' logo, saying: “I enjoy working with Diffsynth-Studio, it's a perfect framework.” "
"The shot starts low—framing her waist, shoes, and a branded floor mat—and smoothly jib-ups upward. "
"As the camera rises, it reveals her smiling face, upper body, and behind her: a bright creative studio with wall art and a large window showing daylight sky. "
"The final frame fully shows the inspiring workspace above the initial view, ensuring spatial continuity."
)
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path='ltx2_camera_jib_up.mp4',
fps=24,
audio_sample_rate=24000,
)

61
examples/ltx2/model_inference/LTX-2-T2AV-Camera-Control-Static.py Normal file
View File

@@ -0,0 +1,61 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.bfloat16,
"offload_device": "cpu",
"onload_dtype": torch.bfloat16,
"onload_device": "cuda",
"preparing_dtype": torch.bfloat16,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Static", origin_file_pattern="ltx-2-19b-lora-camera-control-static.safetensors"),
)
prompt = "A beautiful sunset over the ocean."
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path='ltx2_camera_static.mp4',
fps=24,
audio_sample_rate=24000,
)

62
examples/ltx2/model_inference_low_vram/LTX-2-T2AV-Camera-Control-Dolly-In.py Normal file
View File

@@ -0,0 +1,62 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.float8_e5m2,
"offload_device": "cpu",
"onload_dtype": torch.float8_e5m2,
"onload_device": "cpu",
"preparing_dtype": torch.float8_e5m2,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Dolly-In", origin_file_pattern="ltx-2-19b-lora-camera-control-dolly-in.safetensors"),
)
prompt = "Dolly-in shot: A cheerful girl smiles brightly and says, 'I enjoy working with Diffsynth-Studio, it's a perfect framework.' The camera smoothly moves closer to her face, highlighting her enthusiasm and sincerity."
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path='ltx2_dolly_in_lora.mp4',
fps=24,
audio_sample_rate=24000,
)

62
examples/ltx2/model_inference_low_vram/LTX-2-T2AV-Camera-Control-Dolly-Left.py Normal file
View File

@@ -0,0 +1,62 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.float8_e5m2,
"offload_device": "cpu",
"onload_dtype": torch.float8_e5m2,
"onload_device": "cpu",
"preparing_dtype": torch.float8_e5m2,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Dolly-Left", origin_file_pattern="ltx-2-19b-lora-camera-control-dolly-left.safetensors"),
)
prompt = "Dolly-left shot: A joyful young woman sits at a minimalist desk with a laptop running Diffsynth-Studio, code and generative visuals glowing on screen. She turns slightly toward the camera and says with a smile, 'I enjoy working with Diffsynth-Studio, it's a perfect framework.' As she speaks, the camera smoothly dollies left, revealing a wall of framed open-source project posters, a whiteboard covered in neural network sketches, and a shelf stacked with AI/graphics books beside her."
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path='ltx2_dolly_left_lora.mp4',
fps=24,
audio_sample_rate=24000,
)

63
examples/ltx2/model_inference_low_vram/LTX-2-T2AV-Camera-Control-Dolly-Out.py Normal file
View File

@@ -0,0 +1,63 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.float8_e5m2,
"offload_device": "cpu",
"onload_dtype": torch.float8_e5m2,
"onload_device": "cpu",
"preparing_dtype": torch.float8_e5m2,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Dolly-Out", origin_file_pattern="ltx-2-19b-lora-camera-control-dolly-out.safetensors"),
)
prompt = "Dolly-out shot: A joyful young woman smiles warmly and says: 'I enjoy working with Diffsynth-Studio, it's a perfect framework.' As she speaks, the camera slowly dollies out, revealing a bright, modern creative studio filled with plants, whiteboards full of diagrams, and soft natural light from large windows."
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path=f'ltx2_dolly_out.mp4',
fps=24,
audio_sample_rate=24000,
)

62
examples/ltx2/model_inference_low_vram/LTX-2-T2AV-Camera-Control-Dolly-Right.py Normal file
View File

@@ -0,0 +1,62 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.float8_e5m2,
"offload_device": "cpu",
"onload_dtype": torch.float8_e5m2,
"onload_device": "cpu",
"preparing_dtype": torch.float8_e5m2,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Dolly-Right", origin_file_pattern="ltx-2-19b-lora-camera-control-dolly-right.safetensors"),
)
prompt = "Dolly-right shot: A happy girl looks up and says happily, 'I enjoy working with Diffsynth-Studio, it's a perfect framework.' She sits before a sunlit café table, her open laptop displaying the Github interface. The camera glides right to show a barista crafting coffee in the background, shelves of artisan beans, and a chalkboard menu softly blurred in the bokeh."
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path='ltx2_dolly_right_lora.mp4',
fps=24,
audio_sample_rate=24000,
)

72
examples/ltx2/model_inference_low_vram/LTX-2-T2AV-Camera-Control-Jib-Down.py Normal file
View File

@@ -0,0 +1,72 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.float8_e5m2,
"offload_device": "cpu",
"onload_dtype": torch.float8_e5m2,
"onload_device": "cpu",
"preparing_dtype": torch.float8_e5m2,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Jib-Down", origin_file_pattern="ltx-2-19b-lora-camera-control-jib-down.safetensors"),
)
prompt = (
"A girl is very happy, standing on a clean studio floor with soft ambient lighting. "
"She is speaking directly to the camera: “I enjoy working with Diffsynth-Studio, it's a perfect framework.” "
"The shot begins with a medium close-up framing her from the waist up. As the camera performs a smooth jib-down movement—"
"descending vertically downward—it gradually reveals more of the lower portion of the scene. "
"During the descent, the following elements become visible near the bottom of the frame: "
"- The polished concrete floor with subtle reflections of the girl’s shoes, "
"- A small branded mat labeled “Diffsynth-Studio” placed just beneath her feet, "
"- The lower part of a sleek workstation desk with a glowing logo on its front panel, partially hidden at the start but fully revealed as the camera lowers. "
"This downward motion provides a dynamic reveal of contextual details that reinforce the professional and creative environment, "
"while maintaining focus on the girl’s enthusiastic expression throughout."
)
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path='ltx2_camera_jib_down.mp4',
fps=24,
audio_sample_rate=24000,
)

66
examples/ltx2/model_inference_low_vram/LTX-2-T2AV-Camera-Control-Jib-Up.py Normal file
View File

@@ -0,0 +1,66 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.float8_e5m2,
"offload_device": "cpu",
"onload_dtype": torch.float8_e5m2,
"onload_device": "cpu",
"preparing_dtype": torch.float8_e5m2,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Jib-Up", origin_file_pattern="ltx-2-19b-lora-camera-control-jib-up.safetensors"),
)
prompt = (
"A girl stands happily at a sleek desk with a glowing 'Diffsynth-Studio' logo, saying: “I enjoy working with Diffsynth-Studio, it's a perfect framework.” "
"The shot starts low—framing her waist, shoes, and a branded floor mat—and smoothly jib-ups upward. "
"As the camera rises, it reveals her smiling face, upper body, and behind her: a bright creative studio with wall art and a large window showing daylight sky. "
"The final frame fully shows the inspiring workspace above the initial view, ensuring spatial continuity."
)
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path='ltx2_camera_jib_up.mp4',
fps=24,
audio_sample_rate=24000,
)

61
examples/ltx2/model_inference_low_vram/LTX-2-T2AV-Camera-Control-Static.py Normal file
View File

@@ -0,0 +1,61 @@
import torch
from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelConfig
from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
vram_config = {
"offload_dtype": torch.float8_e5m2,
"offload_device": "cpu",
"onload_dtype": torch.float8_e5m2,
"onload_device": "cpu",
"preparing_dtype": torch.float8_e5m2,
"preparing_device": "cuda",
"computation_dtype": torch.bfloat16,
"computation_device": "cuda",
}
pipe = LTX2AudioVideoPipeline.from_pretrained(
torch_dtype=torch.bfloat16,
device="cuda",
model_configs=[
ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
],
tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
)
pipe.load_lora(
pipe.dit,
ModelConfig(model_id="Lightricks/LTX-2-19b-LoRA-Camera-Control-Static", origin_file_pattern="ltx-2-19b-lora-camera-control-static.safetensors"),
)
prompt = "A beautiful sunset over the ocean."
negative_prompt = (
"blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
"grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
"deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
"wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
"field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
"lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
"valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
"mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
"off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
"pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
"inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
)
height, width, num_frames = 512 * 2, 768 * 2, 121
video, audio = pipe(
prompt=prompt,
negative_prompt=negative_prompt,
seed=43,
height=height,
width=width,
num_frames=num_frames,
tiled=True,
use_two_stage_pipeline=True,
)
write_video_audio_ltx2(
video=video,
audio=audio,
output_path='ltx2_camera_static.mp4',
fps=24,
audio_sample_rate=24000,
)