Do I need specific models installed for this to work?

You need the custom AILab_QwenVL nodes to load successfully; they will rely on Qwen‑VL assets for prompt enhancement. The image and video synthesis nodes (e9d2dda8‑… and 99c27236‑…) use whatever model/checkpoint configuration the workflow provides or you connect. If ComfyUI reports missing models, follow the node warnings to download or point to the correct files.

How can I reduce flicker or subject drift between frames?

Try increasing motion/consistency strength in the video node, lowering noise strength, and keeping a fixed seed in the image stage. Clear, specific prompts from AILab_QwenVL_PromptEnhancer (subjects, camera, style) also help. Shorter clips and moderate resolutions are easier to stabilize during iteration.

What settings should I change for faster previews?

Lower resolution, reduce steps in the image node (e9d2dda8‑…), use a smaller clip length or fps in the video node (99c27236‑…), and keep a fixed seed. Use PreviewAny to validate direction before committing to a full‑quality render.

Can this workflow do image‑to‑video or just text‑to‑video?

Both. You can stop after the image stage and export stills (SaveImage), or feed existing/reference imagery through the prompt and video stages to produce motion. The end‑to‑end path is text‑to‑video, but the stages are modular so you can mix and match.

AI on the Lot 2026 - Final WF - t2v

AI on the Lot 2026 - Final WF - t2v is a complete text-to-video pipeline that stitches together prompt design, image synthesis, and video assembly inside ComfyUI. It uses AILab_QwenVL and AILab_QwenVL_PromptEnhancer to turn a simple idea into a rich, structured prompt. From there, the workflow generates high-quality keyframes with the image synthesis stage (node ID e9d2dda8-2509-4d8d-9702-a9bf00c26650), previews results with PreviewAny, and then builds a cohesive video clip in the video stage (node ID 99c27236-513e-427c-9ed4-49406365f0d3). SaveImage and SaveVideo provide clean, reproducible exports at each stage.

Technically, the flow is: your base text prompt (and optionally a reference image) goes into AILab_QwenVL to extract and enrich scene details. AILab_QwenVL_PromptEnhancer refines the text into a model-friendly prompt with clearer subjects, styles, and shot descriptors. That prompt feeds the image generator node (e9d2dda8-2509-4d8d-9702-a9bf00c26650) to produce one or more seed-controlled keyframes at your chosen resolution. Those frames are then passed to the video node (99c27236-513e-427c-9ed4-49406365f0d3), which synthesizes a clip based on duration and motion/consistency settings. PreviewAny helps you iterate quickly, while SaveImage and SaveVideo lock in stills or a final video once you’re happy. Because the stages are modular, you can use it for text-to-image only, image-to-video only, or end-to-end text-to-video with the same graph.

FAQ

자주 묻는 질문

모든 워크플로우 보기