1019703 — v1.0

EML_LTX_TTV/ITV is a ComfyUI workflow for making LTX-Video clips from a text prompt or uploaded image, with Text To Video and Image To Video modes and WebP or MP4 video output.

In Image To Video mode, the guide says Florence2 supplies the image’s static description, while Llama 3.2 turns your prompt into motion and lighting cues.

For generated images, the guide says CRF adds noise to help address frozen video.

The guide lists prompts and uploaded images as inputs; it saves WebP and MP4 videos, plus temporary PNG images, in `ComfyUI\output`.

If custom nodes are missing at startup, the guide recommends using Manager → Install Missing Custom Nodes, selecting all, installing them, and restarting when prompted.

Suggestion · non vérifié

For Llama 3.2 prompt generation, the guide recommends installing Ollama, running `ollama run llama3.2`, and activating `ltxv` mode.

Suggestion · non vérifié

Listed node packs include ComfyUI_Comfyroll_CustomNodes, comfy-image-saver, ComfyUI-Easy-Use, comfyui-ollama, cg-use-everywhere, ComfyUI-Custom-Scripts, and Comfyui-ergouzi-kaiguan.

Listed model files include `ltx-2b-v0.9-bf16.safetensors` and `t5xxl_fp8_e4m3fn.safetensors`.

The guide lists defaults of 768×512 at 3:2, 25 fps, 97 frames, 30 steps, and CFG 3.0; the aspect-ratio list includes a 16:9 hint.

For adjustments, the guide points to groups marked in blue and advises changing only `steps` in `LTXVScheduler`.

Suggestion · non vérifié

The LTXV Model Configurator is a reference for resolution choices; the guide gives CFG 3.0–3.5 as its recommended range.

Manual prompt mode is in development; the guide says the workflow currently supports LLM input.

The guide advertises use on 8 GB VRAM cards, but says this workflow needs more video memory and is lower quality than its manual workflow.