LTX 2.5 _ text and image to video

🎬 LTX 2.5 – Image to Video (v2) – ComfyUI Workflow

Created by Me – a filmmaker & AI workflow designer. This is the second generation of my image‑to‑video workflow, now powered by LTX 2.5 – the latest open‑source video generation model from Lightricks. It turns a single image into a high‑quality cinematic video with camera movement control, prompt enhancement, two‑stage sampling (low‑res + high‑res), and audio‑aware latent generation – all packed into a clean, modular workflow using my custom nodes.

🧩 What's New in v2

• LTX 2.5 Engine – upgraded from previous LTX versions for better quality, longer videos, and faster generation.

• Two‑Stage Sampling – generates a low‑resolution latent first, then upscales and refines it for higher fidelity.

• Latent Upscaler – uses a dedicated upscale model (ltx-2.5-latent-spatial-upscaler-x2) to double resolution in latent space.

• Audio‑Aware Latents – generates audio latents alongside video for full audiovisual output.

• Camera Movement Selector – my custom node that builds dynamic camera motion prompts (pan, tilt, dolly, orbit, etc.) with two‑phase transitions.

• Negative Prompt Generator – toggles for video‑specific issues like flickering, warping, and temporal artifacts.

• SaveVideoCustom – saves with date, size, seed, and custom folder/project name – fully organised.

• Prompt Enhancer (Gemma 4) – optionally enhances your prompt using a lightweight Gemma 4 model for better detail.

• API Text Encoding (Optional) – use Gemma API to offload text encoding and save VRAM.

This workflow is fully modular and works with my custom node pack – no more node spaghetti!

1️⃣ Required Models & Prerequisites

Make sure these files are placed in your ComfyUI models/ folders:

FolderRequired Filesdiffusion_models/ltx-2.5-22b-distilled-transformer-bf16.safetensorsvae/ltx-2.5-video-vae-bf16.safetensors
ltx-2.5-audio-vae-bf16.safetensorslatent_upscale_models/ltx-2.5-latent-spatial-upscaler-x2-bf16-1.0.safetensorstext_encoders/gemma4-12b-with-proj-ltx-2.5-bf16.safetensors (main encoder)
gemma4_e2b_it_bf16.safetensors (optional enhancer)

⚠️ Installation Required: This workflow uses my custom nodes. You must install (or update) ComfyUI-Saeed-Nodes first:

bash

git clone https://github.com/saeedkhalili/ComfyUI-Saeed-Nodes

Place the cloned folder inside your ComfyUI custom_nodes/ directory and restart ComfyUI.
If you already have it installed, please pull the latest updates – this v2 workflow uses new nodes not available in older versions.

💡 VRAM Requirement: ~16–20 GB recommended for full 2‑stage generation with upscaling. Use the API text encoding option to reduce VRAM usage if needed.

3️⃣ How It Works (Image → Cinematic Video in 5 Steps)

1. Load your image – any image you want to turn into a video.

2. Set camera movement – use my CameraMovementSelector to define a motion path (e.g., "Dolly In through a gate, then Tilt Up to reveal a mountain"). The node supports two‑phase movements for complex shots.

3. Build prompts – write a positive prompt (or let the Gemma 4 enhancer refine it), and generate a negative prompt with the NegativePromptGenerator (with video‑specific toggles).

4. Generate – the workflow:

   • Preprocesses the image (compression, resizing).

   • Creates a low‑resolution latent (8 steps).

   • Upscales the latent using the spatial upscaler.

   • Refines with a second sampler (3 steps) for high‑quality output.

   • Decodes video and audio latents separately.

5. Save – SaveVideoCustom saves the output with your chosen naming conventions (date, size, seed, folder, project name).

All custom nodes are designed for simplicity – just select, toggle, and generate!

6️⃣ Advanced Settings (Optional)

• Camera Movement Selector – define initial view, subject, translation (Dolly, Crane, etc.), rotation (Tilt, Pan, etc.), speed, and a second phase for complex transitions.

• Prompt Enhancer – toggle to use Gemma 4 to refine your positive prompt for better detail and coherence.

• API Text Encoding – use the Gemma API to encode prompts externally, saving VRAM for larger models.

• Two‑Stage Control – adjust noise seed, CFG, and sampler for both the low‑res and high‑res stages independently.

• Duration & FPS – set video length and frame rate (default: 5 sec @ 24 fps).

• Image Compression – control the compression level of the input image before VAE encoding.

• Bit Depth – choose 8‑bit or 16‑bit video output.

• Save Customisation – include date, resolution, seed, and project name in the filename.

📦 Final Output

After running, you'll get a high‑quality cinematic video (with audio) saved in your output folder with a descriptive filename – ready for editing, social media, or further processing.

Why I built this: As a filmmaker, I needed a powerful, flexible image‑to‑video tool that could handle complex camera movements and produce broadcast‑quality results. LTX 2.5 delivers that, and my custom nodes make it easy to control every aspect of the shot. Whether you're storyboarding, creating concept videos, or producing final content, this workflow gives you the power of a professional VFX tool inside ComfyUI. I hope it takes your projects to the next level! 🎥✨

📬 Contact: If you’d like to donate, ask a question, or order a custom workflow, reach out to me on Civitai (SaeedKhalili), visit my website SaeedKhalili.com, or email me at [email protected]. I’m always happy to help fellow creators!