This workflow uses SAM to invert the colors of the character to be replaced. By doing so, it makes character replacement significantly more consistent that Minimax H3 alone.<h1 id="how-to-use-it">How to use it</h1>You must<ol><li>(v2 and above) Install <a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-SolAttn_triton">https://github.com/kijai/ComfyUI-SolAttn_triton</a> if you want the speedup.</li><li>(v3) Install <a target="_blank" rel="ugc" href="https://github.com/1038lab/ComfyUI-QwenVL">https://github.com/1038lab/ComfyUI-QwenVL</a> for QwenVL video to text description.</li><li>(v1 only) Change the text prompt to describe both the scenario and the subject. Your goal is that Minimax H3 reproduces the original video as similar as possible.</li><li>Change the video and the picture, and make sure to change as well the resolution to match the original video.</li><li>(Optional, v2) change the SAM prompt until it masks only the subject to replace.</li><li>(Optional, v3) change the hint given to the Qwen node. </li></ol><h3 id="why-sam-and-masking-with-h3">Why SAM and masking with H3</h3>When using H3, it seems that the model will either recreate the original video as it is, without changing the subject at all (follows the &lt;Video 1&gt; too literally), or it will create a totally new video (follows the &lt;Picture 1&gt; too literally).However, when inverting the color of the subject in the video, H3 is forced to produce a new video because the subject looks weird. So it can't follow the &lt;Video 1&gt; too literally anymore, otherwise it would produce a video with inverted colors.

Minimax H3 character replacement

With QwenVL, describes the scene and the subject to do the replacement

Files with SHA256: 3e5f6eee46ad2d7ab9ced4128b5a5fcffc2d7c4c4723254251109f3a936f4cc3