ComfyUI and LTX 2.3: cartoon workflow tutorial

This Russian-language tutorial follows a short cartoon project through image preparation, animation and sound design in ComfyUI. The speaker uses LTX 2.3 with a relatively simple workflow and generates the main cartoon at 512x704. He explains a six-step sampling setup, image resizing to dimensions divisible by 32, and switches for selecting image, text and audio inputs. His results illustrate local AI animation on his own computer, without establishing a hardware minimum or a general speed benchmark.

Reference preparation takes up much of the lesson. The speaker selects images with similar lighting, removes rain through his Client 9B panel, and crops or extends backgrounds to fit the chosen frame size. He demonstrates approximate color matching in the panel, acknowledges its imperfect results, and shows Photoshop's Match Color and masks as another approach.

Gemma 4 turns scene descriptions into animation prompts. In his examples, it writes the instructions in English while retaining Russian dialogue. A separate CosyVoice demonstration uses a voice sample with either typed text or recorded speech; he says he did not use externally generated voices in the opening cartoon.

The final sections connect audio to the animation workflow and add Stable Audio 3 effects such as bicycle sounds and footsteps. The speaker notes that the first sound generation takes longer while models load. He also discusses Topaz Video for enlarging and smoothing the small output, without giving a detailed settings tutorial.