AI on the Lot 2026_LTX-2.3_IC LoRA (Union Control)

What's worth including here isn't the final video, it's the pattern of using an LLM as a prompt intermediary. Two reference images and a short description go into GeminiNode, which expands them into a detailed prompt in the format Seedance 2.0 responds well to; that generated prompt, together with the reference images, then feeds ByteDance2ReferenceNode for the actual render, with a MarkdownNote in between that lays the LLM-written prompt out for a human to read and edit. For anyone who keeps re-tuning prompt formatting for the same video model, this two-stage LLM-assist-then-generate structure is directly reusable.

This workflow is not flashy — it solves a real tattoo-design communication problem. A PrimitiveStringMultiline node takes a one-line idea, a ComfySwitchNode toggles between common tattoo styles such as blackwork, fine-line and geometric, StringConcatenate assembles the final prompt, and a ResolutionSelector presets canvas ratios for patch, sleeve or flash-sheet layouts. The result is a guided input flow rather than a single block of prompt text, which is why it's worth keeping around — the same style-switch-plus-canvas-preset pattern transfers directly to other stylized image generation needs.

A highly detailed video prompt for Seedance 2.5 mimicking a handheld 16mm analog morning vlog of a woman walking her dog, focusing on realistic imperfections and film texture.