<h2 id="version-2.0-notes:">Version 2.0 notes:</h2><ul><li>Added an optional 2nd pass for upscale refinement</li><li>More ways to save images from the 90 frame generated clip</li><li>Additional sparsity node for Eros Max usage</li></ul><h2 id="h3-img2img-character-swap-workflow">H3 IMG2IMG Character Swap Workflow</h2>This ComfyUI workflow uses MiniMax H3 and the <a target="_blank" rel="ugc" href="https://huggingface.co/Alissonerdx/Minimax-H3-ComfyUI/tree/main/loras">NRDX Character Swap LoRA</a> for identity-guided image generation. It combines head and body reference images into a character reference and uses a separate pose image to guide the generated result.Swap out just a head, body, clothing, or an entire character.Experiment by adding two or more character references to the model to replace multiple characters in an image.Insert an entirely new character into a scene.The workflow is intended for experimentation, creative production, visual effects, research, and other lawful uses where users have the necessary rights and permissions to use their source media and reference identities.<h3 id="current-workflow-configuration">Current Workflow Configuration</h3>As provided, this workflow is configured to generate a batch of 90 frames. It repeats the pose image as a guide and allows you to save frames from the generated result.<h3 id="important-usage-notes">Important Usage Notes</h3>Results can vary depending on the reference images, pose, framing, resolution, prompt, seed, and other generation settings.The 90 frames is an arbitrary choice but I've received good results from it.The seed can make a substantial difference to identity fidelity, facial structure, body proportions, clothing, blending, and artifact levels. If a result is poor, test multiple seeds before making extensive changes to the workflow.Clear reference images generally give the model more useful information. Small faces, obscured features, extreme angles, inconsistent reference images, or low-resolution inputs can reduce accuracy. Higher working resolutions may improve detail but require more VRAM and processing time.The generated result may change the subject’s appearance, pose, clothing, or surrounding scene in unintended ways. The workflow does not guarantee exact identity preservation, accurate anatomy, or artifact-free output.<h3 id="required-models-and-custom-nodes">Required Models and Custom Nodes</h3>The workflow requires MiniMax H3 models, video and audio VAEs, a text encoder, LoRAs, and ComfyUI custom nodes. Optional branches require additional components for video output, attention acceleration, or upscaling.A dependency list, model filenames, setup instructions, and available download sources are included in the workflow’s Notes section. Review that section before reporting missing-node or missing-model errors.Model availability, licensing, compatibility, and download locations may change. Users are responsible for reviewing and complying with the licenses and terms associated with each third-party model, custom node, repository, and dependency.<h3 id="consent-and-approval-requirements">Consent and Approval Requirements</h3>By downloading or using this workflow, you agree to use images, likenesses, identities, and other personal media only when you have the appropriate authorization, consent, license, or legal right to do so.Obtain approval from identifiable individuals before creating or distributing synthetic or altered media depicting them, particularly when the result could reasonably be mistaken for an authentic image.Do not use this workflow to impersonate another person, create fraudulent or deceptive material, bypass identity-verification systems, harass or exploit individuals, or otherwise violate applicable laws, platform rules, privacy rights, publicity rights, intellectual-property rights, or contractual obligations.If source or reference material belongs to a client, employer, performer, photographer, studio, production company, or another third party, you are responsible for confirming that your authorization covers AI-assisted modification and redistribution.<h3 id="user-responsibility">User Responsibility</h3>You are solely responsible for:<ul><li>The media you provide to the workflow.</li><li>Obtaining necessary permissions, licenses, and releases.</li><li>Reviewing generated outputs before publication or distribution.</li><li>Determining whether disclosure of AI-generated or AI-modified content is required.</li><li>Following the licenses associated with the models and custom nodes used by the workflow.</li><li>Complying with applicable laws, regulations, contractual requirements, and platform policies.</li><li>Ensuring that generated media is not used deceptively, unlawfully, or without appropriate authorization.</li></ul><h3 id="liability-disclaimer">Liability Disclaimer</h3>This workflow is provided “as is” without warranties or guarantees of any kind.The workflow does not guarantee accuracy, identity fidelity, compatibility, performance, reliability, fitness for a particular purpose, or continued compatibility with future versions of ComfyUI, custom nodes, models, drivers, or hardware.The workflow’s creator is not responsible for any loss, damage, claim, legal dispute, account action, copyright or licensing issue, privacy complaint, reputational harm, hardware problem, data loss, or other consequence arising from the installation, modification, distribution, or use of this workflow or its outputs.Third-party models, nodes, repositories, software, and services remain the responsibility of their respective creators and are subject to their own licenses and terms.<h3 id="download-agreement">Download Agreement</h3>By downloading, importing, modifying, or using this workflow, you acknowledge that:<ol><li>You understand that this workflow can alter a person’s apparent identity in generated media.</li><li>You will obtain appropriate consent or authorization for identifiable people depicted in your source or reference material.</li><li>You will not use the workflow for fraud, impersonation, deception, harassment, exploitation, or other unlawful activity.</li><li>You accept responsibility for the media you create and distribute.</li><li>You will review and comply with the licenses and requirements of all referenced third-party models and custom nodes.</li><li>You understand that results are experimental and may contain visual errors or unintended changes.</li><li>You accept that the workflow is provided without warranty and that use is at your own risk.</li></ol>Downloading or using this workflow constitutes acceptance of these conditions.

Minimax H3 IMG2IMG Character Swap+ Workflow

Version 2.0 Includes:<ul><li>An additional Two Pass option for refinement</li><li>0.5 MP first pass → 1 MP latent upscale → H3 refinement </li><li>Try 1. MP first pass and then <a target="_blank" rel="ugc" href="http://2.MP">2. MP</a> upscale.</li><li>Second-pass defaults: 3 steps, 0.30 denoise</li><li>A Skip Second Pass switch and optional first-pass comparison outputs</li></ul><ul><li>Frame selection before saving, with optional SeedVR2 afterward</li><li>Installation and testing instructions in Notes</li></ul>Requires the H3 Latent Upscaler extension and model listed in Notes.<ul><li>Keep the second-pass group active even when skipping refinement—it contains the output switch and decoder. just use the red boolean node to skip.</li><li>Both FL2VA and Ref2VA models seem to work. test and see if you find any significant differences.</li></ul>

Files with SHA256: 21ed9e6c03de458160a0b6817d6fbd9bd83e307892de9b0f7a22a1c6c1d4ccb1