Add support for I2V during validation

configs/ltxv_2b_full.yaml

@@ -38,6 +38,7 @@ validation:
     - "a professional portrait video of a person with blurry bokeh background"
     - "a video of a person wearing a nice suit"
   negative_prompt: "worst quality, inconsistent motion, blurry, jittery, distorted"
+  images: null # Set to a list of image paths to use first-frame conditioning, or null to disable
   video_dims: [768, 448, 89] # [width, height, frames]
   seed: 42
   inference_steps: 50

configs/ltxv_2b_lora.yaml

@@ -49,6 +49,7 @@ validation:
     - "a professional portrait video of a person with blurry bokeh background"
     - "a video of a person wearing a nice suit"
   negative_prompt: "worst quality, inconsistent motion, blurry, jittery, distorted"
+  images: null # Set to a list of image paths to use first-frame conditioning, or null to disable
   video_dims: [768, 448, 89] # [width, height, frames]
   seed: 42
   inference_steps: 50

configs/ltxv_2b_lora_low_vram.yaml

@@ -49,6 +49,7 @@ validation:
     - "a professional portrait video of a person with blurry bokeh background"
     - "a video of a person wearing a nice suit"
   negative_prompt: "worst quality, inconsistent motion, blurry, jittery, distorted"
+  images: null # Set to a list of image paths to use first-frame conditioning, or null to disable
   video_dims: [768, 448, 89] # [width, height, frames]
   seed: 42
   inference_steps: 50

configs/ltxv_2b_lora_template.yaml

@@ -48,6 +48,7 @@ validation:
     - "a professional portrait video of a person with blurry bokeh background"
     - "a video of a person wearing a nice suit"
   negative_prompt: "worst quality, inconsistent motion, blurry, jittery, distorted"
+  images: null # Set to a list of image paths to use first-frame conditioning, or null to disable
   video_dims: [768, 448, 89] # [width, height, frames]
   seed: 42
   inference_steps: 30

src/ltxv_trainer/config.py

@@ -189,6 +189,12 @@ class ValidationConfig(ConfigBaseModel):
         description="Negative prompt to use for validation examples",
     )
 
+    images: list[str] | None = Field(
+        default=None,
+        description="List of image paths to use for validation. "
+        "One image path must be provided for each validation prompt",
+    )
+
     video_dims: tuple[int, int, int] = Field(
         default=(704, 480, 161),
         description="Dimensions of validation videos (width, height, frames)",

src/ltxv_trainer/trainer.py

@@ -11,7 +11,7 @@
 import torch
 from accelerate import Accelerator
 from accelerate.utils import set_seed
-from diffusers import LTXPipeline
+from diffusers import LTXImageToVideoPipeline, LTXPipeline
 from diffusers.utils import export_to_video
 from loguru import logger
 from peft import LoraConfig, get_peft_model_state_dict
@@ -49,7 +49,7 @@
 from ltxv_trainer.model_loader import load_ltxv_components
 from ltxv_trainer.quantization import quantize_model
 from ltxv_trainer.timestep_samplers import SAMPLERS
-from ltxv_trainer.utils import get_gpu_memory_gb
+from ltxv_trainer.utils import get_gpu_memory_gb, open_image_as_srgb
 
 # Disable irrelevant warnings from transformers
 os.environ["TOKENIZERS_PARALLELISM"] = "true"
@@ -645,13 +645,30 @@ def _sample_videos(self, progress: Progress) -> list[Path] | None:
         if not self._config.acceleration.load_text_encoder_in_8bit:
             self._text_encoder.to(self._accelerator.device)
 
-        pipeline = LTXPipeline(
-            scheduler=deepcopy(self._scheduler),
-            vae=self._vae,
-            text_encoder=self._text_encoder,
-            tokenizer=self._tokenizer,
-            transformer=self._transformer,
-        )
+        use_images = self._config.validation.images is not None
+
+        if use_images:
+            if len(self._config.validation.images) != len(self._config.validation.prompts):
+                raise ValueError(
+                    f"Number of images ({len(self._config.validation.images)}) must match "
+                    f"number of prompts ({len(self._config.validation.prompts)})"
+                )
+
+            pipeline = LTXImageToVideoPipeline(
+                scheduler=deepcopy(self._scheduler),
+                vae=self._vae,
+                text_encoder=self._text_encoder,
+                tokenizer=self._tokenizer,
+                transformer=self._transformer,
+            )
+        else:
+            pipeline = LTXPipeline(
+                scheduler=deepcopy(self._scheduler),
+                vae=self._vae,
+                text_encoder=self._text_encoder,
+                tokenizer=self._tokenizer,
+                transformer=self._transformer,
+            )
         pipeline.set_progress_bar_config(disable=True)
 
         # Create a task in the sampling progress
@@ -665,25 +682,33 @@ def _sample_videos(self, progress: Progress) -> list[Path] | None:
 
         video_paths = []
         i = 0
-        for prompt in self._config.validation.prompts:
+        for j, prompt in enumerate(self._config.validation.prompts):
             generator = torch.Generator(device=self._accelerator.device).manual_seed(self._config.validation.seed)
 
             # Generate video
             width, height, frames = self._config.validation.video_dims
+
+            pipeline_inputs = {
+                "prompt": prompt,
+                "negative_prompt": self._config.validation.negative_prompt,
+                "width": width,
+                "height": height,
+                "num_frames": frames,
+                "num_inference_steps": self._config.validation.inference_steps,
+                "generator": generator,
+            }
+
+            if use_images:
+                image_path = self._config.validation.images[j]
+                pipeline_inputs["image"] = open_image_as_srgb(image_path)
+
             with autocast(self._accelerator.device.type, dtype=torch.bfloat16):
-                videos = pipeline(
-                    prompt=prompt,
-                    negative_prompt=self._config.validation.negative_prompt,
-                    width=width,
-                    height=height,
-                    num_frames=frames,
-                    num_inference_steps=self._config.validation.inference_steps,
-                    generator=generator,
-                ).frames
+                result = pipeline(**pipeline_inputs)
+                videos = result.frames
 
             for video in videos:
                 video_path = output_dir / f"step_{self._global_step:06d}_{i}.mp4"
-                export_to_video(video, str(video_path), fps=24)
+                export_to_video(video, str(video_path), fps=25)
                 video_paths.append(video_path)
                 i += 1
             progress.update(task, advance=1)