Save 4 hours! We auto-setup your workflow! Free!

Drop your workflow.json — we handle every dependency, custom node, and model. Just open the link and run.

Auto-Setup Workflow Json (Free) Now!
ComfyUI > Nodes > H3 Character Sheet Designer

ComfyUI Extension: H3 Character Sheet Designer

Repo Name

H3-Character-Sheet-Designer

Author
ukr8b3g-cmyk (Account age: 400 days)
Nodes
View all nodes(1)
Latest Updated
2026-11-01
Github Stars
0.41K

How to Install H3 Character Sheet Designer

Install this extension via the ComfyUI Manager by searching for H3 Character Sheet Designer
  • 1. Click the Manager button in the main menu
  • 2. Select Custom Nodes Manager button
  • 3. Enter H3 Character Sheet Designer in the search bar
After installation, click the Restart button to restart ComfyUI. Then, manually refresh your browser to clear the cache and access the updated list of nodes.

Visit ComfyUI Online for ready-to-use ComfyUI environment

  • Free trial available
  • 16GB VRAM to 80GB VRAM GPU machines
  • 400+ preloaded models/nodes
  • Freedom to upload custom models/nodes
  • 200+ ready-to-run workflows
  • 100% private workspace with up to 200GB storage
  • Dedicated Support

Run ComfyUI Online

H3 Character Sheet Designer Description

H3 Character Sheet Designer creates character reference sheets in ComfyUI using MiniMax H3, combining a main image, optional detail references, and up to 13 selected views with customizable layouts.

H3-Character-Sheet-Designer Introduction

H3-Character-Sheet-Designer helps you create character reference sheets in ComfyUI using MiniMax H3’s native image-reference capabilities. You provide a character image, choose the views you need, and optionally add references for details such as a face, outfit, hands, or footwear.

A character sheet brings several views of a design together in one image. This can help you review a character’s appearance, communicate an outfit design, or prepare visual references for further artwork.

Current package version: v1.4.1

H3-Character-Sheet-Designer interface

The extension supports:

  • One main character image and up to three optional part-reference images.
  • Up to 13 views: six head views, three full-body views, front and back clothing views, one paired-hands view, and one paired-feet/footwear view.
  • View presets, including Standard · 5 views and 7 views · hands and feet.
  • An optional layout image to help guide the arrangement.
  • Automatic or manually specified output dimensions.
  • Part instructions and reference roles to describe which details should come from each image.
  • Editable base instructions for individual views in the supported prompt editor.
  • Current MiniMax H3 Image and compatibility workflows for earlier H3 model combinations.

The Designer helps organize references, layout instructions, and view requests. It does not guarantee an exact turnaround or mechanically accurate reconstruction. Identity, unseen details, clothing, and viewpoint consistency still depend on the image model and generation settings.

How H3-Character-Sheet-Designer Works

Think of the Designer as an art director preparing a brief for an image model:

  1. Your main image establishes the character.
    It provides the baseline appearance the model should follow.

  2. Optional references clarify selected details.
    For example, you can provide a separate outfit reference or an image showing the intended footwear.

  3. Your view choices define the sheet’s contents.
    You can request a single depiction, a smaller selection of views, or all 13 supported views.

  4. The Designer combines the instructions.
    It prepares the text prompt, output dimensions, reference-role instructions, and an optional layout guide.

  5. H3 generates the sheet.
    The image model interprets these instructions and references to create a new image.

Direct image references, not an extra image-to-text step

The main character image and three optional part images pass through the Designer unchanged. H3 receives them through its native image-reference conditioning—the mechanism that lets a supplied image influence generation.

The Designer does not introduce an additional language model or a separate text-generation stage to describe your character first.

What the layout image does

The layout image is a visual guide to the requested arrangement, similar to a rough page plan.

  • Layout Image OFF: the guide is not sent to H3. The model receives the textual layout instructions.
  • Layout Image ON: the guide is also supplied as an image reference.

The interface displays the same guide arrangement and dimensions in either mode. Seeing the guide on screen does not mean it is being used as a generation reference.

Start with Layout Image OFF, then try ON if the composition needs help. A guide can improve arrangement, but the model may also copy guide content or adopt its proportions.

How reference priorities work

In v1.4.1, the intended priority is:

  1. Explicit settings, such as bare hands or no shoes.
  2. Part prompts compatible with those settings.
  3. Dedicated part references.
  4. The main character image as the baseline.

If dedicated references with the same role conflict, the prompt favors the later input.

For example, if the main image shows gloves but you explicitly choose bare hands, the instructions prioritize bare hands. If you choose no shoes, the prompt asks to retain reference socks or tights rather than automatically requesting bare feet.

These priorities are expressed through text instructions. They are not isolated image-editing masks, so details can still spill into unintended panels.

H3-Character-Sheet-Designer Features

Select the views you need

View categoryAvailable viewsWhat it helps you examine
Head6Head, face, and hair from the supported viewpoints
Full body3Overall silhouette, proportions, and clothing
Clothing2Front and back outfit appearance
Hands1 paired-hands viewHand details and relevant accessories
Feet/footwear1 paired-feet viewFeet, socks, tights, or shoes

The maximum selection is 13 views in one sheet, not 13 separately saved images.

Useful arrangements include:

  • Single view: the MiniMax H3 Image workflow requests one centered depiction.
  • Standard · 5 views: the default preset in the MiniMax H3 Image workflow.
  • 7 views · hands and feet: two portraits stacked on the left, three full-body views in the center, and hands above feet on the right.
  • All 13 views with Layout Image OFF: six portraits on the left, three full-body views in the center, front/back clothing next, and hands above footwear at the far right.
  • Only hands and footwear with Layout Image OFF: v1.4.1 requests one hands close-up above one feet close-up, separated by white space.

These are requested arrangements, not guaranteed panel counts or positions.

Add references with clear roles

You can assign the optional reference images these roles:

  • Same character
  • Face
  • Head/face/hair
  • Complete outfit
  • Outfit front
  • Outfit back
  • Hands
  • Footwear

For example, you could use the main image for the character, a second image for the complete outfit, and a third image for footwear.

Use Part prompts to clarify your intent. These instructions help explain what a reference is allowed to change, but they do not strictly confine its influence to one panel.

With Auto, unspecified appearance is intended to remain consistent with the references.

Control hands, footwear, and clothing

Bare hands and no shoes

Version 1.4.1 adjusts the default hand and footwear instructions when these explicit settings are selected. This avoids asking the model both to preserve a reference item and remove it.

  • Bare hands replaces conflicting default hand-preservation text.
  • No shoes replaces conflicting footwear-preservation text while retaining reference socks or tights.
  • Auto and user-edited view text keep their existing behavior.

Clothing shape

The clothing-shape choices are:

  • Auto
  • Pants
  • Skirt or robe

Choose the option that describes the outfit you want represented. For example, Skirt or robe provides a more appropriate shape instruction for a long robe than Pants.

Clothing panels use garment-extraction instructions, including flat-outfit and No human wording. Clothing-only output focuses on outfit features rather than preserving the person’s identity. However, bodies or human remnants can still appear.

The older Worn/Garments-only switch was removed. In v1.3, saved Worn values are interpreted as garment extraction.

Choose automatic or manual dimensions

The MiniMax H3 Image workflow defaults to Auto / High detail.

Resolution choiceMeaning
Auto / CurrentOne of the supplied automatic resolution choices
Auto / High detailThe default automatic resolution choice
Auto / 8 MPA higher-pixel-count automatic option
Auto / 16 MPA higher-pixel-count automatic option
ManualUses the width and height you enter

The source documentation does not specify exact dimensions for every automatic choice.

MiniMax H3 Image automatic resolution menu

Automatic layouts can become wider to give full-body figures enough room without changing their proportions. This can increase generation time and graphics memory usage.

Manual preserves your chosen canvas size, but a narrow canvas may leave limited room for many views.

13-view MiniMax H3 Image example with Layout Image OFF

This 4832 × 1280 example produced each requested view once. It is an example result, not a guarantee. Reference images, seed, and resolution can affect detail and viewpoint adherence.

Edit individual view instructions

Package v1.3.1 added a Prompts tab to the H3 Character Sheet Designer v1.3.0 node.

To use the documented editor:

  1. Open Prompts, after Settings.
  2. Select one of the 13 views.
  3. Edit its H3 base instruction.
  4. Select Apply to save the override.
  5. Use Reset this prompt to restore that view’s default instruction.

Important details:

  • You edit one view’s base instruction, not the entire final prompt.
  • Layout instructions, references, Part prompts, and settings are added separately.
  • Use H3 instructions rather than Anima tags.
  • Blank overrides are allowed.
  • Each override accepts up to 2,000 UTF-16 units; this is a text-storage limit, not necessarily 2,000 visible characters.
  • Valid pending edits are also committed when you save or queue the workflow.
  • Overrides persist in saved workflows.
  • Unedited workflows retain their original instructions.
  • The older v1.2.2 node does not have this editor.

Keep viewpoint requirements clear when editing. Custom text can weaken the model’s adherence to the requested angle.

Use the supplied generation settings as a starting point

The current MiniMax H3 Image template uses:

SettingSupplied value
SamplerEuler
SchedulerSimple
Steps6
CFG1
Denoise1
Save nodeComfyUI’s built-in Save Animated WEBP
Save frame rate1 FPS

Although the save node includes “Animated” in its name, this template generates one static image.

For comparisons, keep the reference images and seed fixed. A seed is the number controlling the generation’s starting randomness; keeping it unchanged makes setting comparisons more useful.

H3-Character-Sheet-Designer Models and Workflows

The Designer is not itself an image-generation model. It prepares instructions and references for the model combination used by your workflow.

Keep each workflow’s model components together. The current MiniMax H3 Image, P-H3 compatibility, and older Pika templates use different combinations.

Current MiniMax H3 Image workflow

Use this template for the new MiniMax H3 Image workflow:

MiniMax H3 Image Character Sheet Designer

  • Designer node: MiniMax H3 Image Character Sheet Designer v1.4.0
  • Requires Designer v1.4.0+ and ComfyUI 0.45.0+
  • Uses MiniMax H3 Image Conditioning
  • CLIP loader type: minimax
ComponentRequired model file
Diffusion model—the main image generatorminimax_h3_image_pruned_int8_convrot.safetensors
Text encoder—interprets the prompt for the modelqwen3vl_32b_minimax_h3_nvfp4_awq.safetensors
VAE—converts the model’s internal image representation into a viewable imageminimax_h3_image_vae_fp16.safetensors

Use the exact three files selected in the template. See ComfyUI’s MiniMax H3 Image guide for model-specific guidance.

P-H3 compatibility workflow

Use this template when continuing with the supplied P-H3 workflow:

H3 Character Sheet Designer — P-H3 Subgraph

  • Designer node: H3 Character Sheet Designer v1.3.0
  • Uses P-H3 Image Conditioning
  • Keeps its original node sockets and behavior
ComponentModel file
Diffusion modelminimax_h3_fl2va_pruned_int8_convrot.safetensors
Text encoderp-h3-image_text_encoder_int8_convrot.safetensors
VAEp-h3-image_vae_int8_convrot.safetensors
LoRA adapterp-h3-image_lora_bf16.safetensors

Older Pika v1.3 workflow

The archived v1.3 workflow uses Hybrid V4, an H3 text encoder, and Pika Ref2VA.

ComponentSupplied choice or documented alternative
Template diffusion modelDaSiWa Hybrid V4 INT8 (https://civitai.red/models/2877206/dasiwa-minimax-h3?modelVersionId=3391862) — DasiwaMinimaxH3_dasiwaHybridV4_int8.safetensors
Native diffusion alternativeminimax_h3_ref2va_pruned_int8_convrot.safetensors
H3 INT8 text encoderqwen3vl_32b_minimax_h3_int8_convrot.safetensors
Alternative H3 text encoderqwen3vl_32b_minimax_h3_nvfp4_awq.safetensors
H3 video VAEminimax_h3_video_vae_int8_convrot.safetensors
Alternative H3 video VAEminimax_h3_video_vae_fp16.safetensors
Enabled Pika adapterpika_h3_flex_ref2va_v1_bf16.safetensors
Disabled Pika adapterpika_h3_flex_fl2va_v1_bf16.safetensors

For this older template:

  • Set CLIP type to minimax and use an H3 encoder.
  • Pika Ref2VA is enabled at 1.0; FL2VA is disabled.
  • Match the adapter to the base model and conditioning path.
  • The workflow uses eight ManualSigmas intervals, samples five frames, and saves the first decoded frame as the static sheet.
  • It requires native ComfyUI H3/subgraph support and rgthree-comfy for Power LoRA Loader.

These are supplied comparison settings, not universally optimal settings.

The author has not provided a controlled comparison establishing which model combination offers the best quality, speed, or graphics-memory usage. Follow each model’s separate license and access requirements.

What’s New with H3-Character-Sheet-Designer

v1.4.1 — Clearer hands, footwear, and reference priorities

  • Bare hands and no shoes: explicit choices replace conflicting preservation text in the default views. No shoes retains reference socks or tights.
  • Hands-and-feet layout: selecting only these views with Layout Image OFF requests hands above feet with white space between them.
  • Reference priorities: explicit settings come first, followed by compatible Part prompts and dedicated references. Later references take priority when the same role conflicts.
  • Compatibility preserved: workflow files and node names remain unchanged, including the P-H3 workflow.

A fixed-reference, fixed-seed GPU comparison produced one bare-hands view and one view of feet in black socks with a white gap. This demonstrates one result, not consistent success across all inputs.

Restart ComfyUI after updating.

v1.4.0 — MiniMax H3 Image workflow

  • Added the dedicated MiniMax H3 Image Character Sheet Designer v1.4.0 node and template.
  • Added the default combination Standard · 5 views / Auto · High detail / Layout Image OFF.
  • Added Auto choices Current, High detail, 8 MP, and 16 MP, while retaining Manual dimensions.
  • Replaced numeric panel-coordinate instructions with shorter image-specific layout descriptions.
  • Added a centered single-view request and a descriptive all-13 arrangement.
  • Supplied Euler / Simple sampling at six steps, CFG 1, and denoise 1.
  • Used the built-in Save Animated WEBP node to save a static result.
  • Retained the P-H3 v1.3.0 node and the deprecated v1.2.2 node for saved-workflow compatibility.
  • Provided direct downloads for both current templates.

v1.3.1 — Editable view prompts

  • Added the per-view Prompts editor to the v1.3.0 node.
  • Added Apply, individual reset, blank overrides, and saved-workflow persistence.
  • Preserved existing node names, sockets, image pass-through, and Layout Image behavior.
  • Left unedited workflows’ instructions unchanged.
  • Did not add the editor to v1.2.2.

The author reported 176 Python tests and 112 JavaScript/DOM tests passing, plus live interface and reference-numbering checks. These checks validate implementation behavior, not image quality or performance.

v1.3 — Shared layout guide and expanded controls

  • Provided versioned v1.3.0 and v1.2.2 nodes.
  • Reordered presets to follow the MOSS-style order and added 7 views · hands and feet.
  • Unified the displayed guide and dimensions for Layout Image ON and OFF.
  • Reduced guide margins and gaps.
  • Widened Auto full-body columns, potentially increasing time and graphics-memory requirements.
  • Added compact controls and a Settings tab.
  • Removed the Worn/Garments-only switch while retaining clothing-shape choices.
  • Adjusted clothing prompts toward flat garment extraction and No human instructions.
  • Supplied the older Hybrid V4 / H3 INT8 encoder / Pika Ref2VA subgraph template.

These clothing and layout changes did not demonstrate a major general improvement for clothing-only or all-13 results.

Earlier workflow compatibility

The v1.2.2 node remains available for compatible saved workflows, but is deprecated in the current package.

Earlier Designer, Reference, Multi Reference, and Reference V1.1 IDs were not registered in v1.3. Workflows requiring those IDs may need an appropriate older release. Saved workflows and custom node titles are not automatically rewritten.

See the CHANGELOG and v1.3 release notes for historical details.

Troubleshooting H3-Character-Sheet-Designer

“MiniMax H3 Image Conditioning” is missing

  1. Confirm that you are opening the MiniMax H3 Image template, rather than the P-H3 or older Pika template.
  2. Check that ComfyUI is 0.45.0 or newer.
  3. Check that the Designer package is v1.4.0 or newer.
  4. After updating, restart ComfyUI and reload the browser.

The interface still looks unchanged after an update

  1. Restart ComfyUI.
  2. Reload the browser; use Ctrl+F5 for a full refresh.
  3. Check the node’s displayed version.

Package and node versions do not always match. For example, package v1.4.1 still uses the MiniMax H3 Image Character Sheet Designer v1.4.0 display name, while the P-H3 node retains v1.3.0.

A sampler reports a matrix-shape error

An incompatible text encoder can cause errors such as a mismatch between 32768 and 5120.

  1. Identify which template you are using.
  2. Compare its selected model, text encoder, VAE, and adapter with the corresponding table above.
  3. Set the CLIP loader type to minimax.
  4. For the older Pika template, select an H3 encoder, not a Music3 encoder.

A connected Load Image node is empty or invalid

  1. Select a valid character image in the main Load Image node.
  2. Check every connected optional Load Image node.
  3. Select valid images for those inputs, or disconnect unused optional inputs from the Designer.
  4. Queue the workflow again.

Model files and reference images are not bundled with the templates.

Views are missing, duplicated, or facing the wrong direction

  1. Check the selected views and preset.
  2. If you edited view prompts, restore the affected defaults with Reset this prompt.
  3. Keep the references and seed fixed while comparing one setting at a time.
  4. Compare Layout Image OFF with ON to see whether the guide helps.
  5. Review whether your chosen dimensions give the requested views enough room.

There is no guaranteed correction. Missing views, duplicates, and imperfect angles remain known limitations.

Guide shapes or unwanted proportions appear in the result

  1. Set Use Layout Image to OFF.
  2. Keep the guide wire connected in the older Pika workflow; OFF already outputs None for the guide.
  3. Generate again using the same seed and references to compare.

The interface will still show its layout preview. That is expected.

Clothing panels contain a person or body parts

  1. Check the clothing-shape setting.
  2. Review Part prompts and references for conflicting instructions.
  3. If you customized the clothing view text, compare it with the default extraction instructions.
  4. Keep in mind that No human and flat-outfit wording are requests, not guaranteed removal tools.

Human remnants and confusion between clothing and full-body views remain possible.

Bare hands or no shoes are not being respected

  1. Check that you are using package v1.4.1 and have restarted ComfyUI.
  2. Select the explicit bare hands or no shoes setting.
  3. Review Part prompts for incompatible instructions.
  4. If you edited the hand or footwear base prompt, review it or reset it: custom view text retains its existing behavior.
  5. Check dedicated references for conflicting details.

No shoes does not mean barefoot: reference socks and tights are intentionally retained.

Generation is slow or runs out of graphics memory

  1. Check whether you selected 8 MP or 16 MP.
  2. Try a less demanding resolution choice or smaller Manual dimensions.
  3. Consider whether you need all 13 views for the current comparison.
  4. Remember that Auto layouts can grow wider to accommodate full-body panels.

First model loading or switching models is separate from the Designer’s layout preparation. No controlled speed or memory benchmark is provided.

An older workflow has missing nodes or broken connections

  1. Identify whether it is the current MiniMax H3 Image, P-H3, or archived Pika workflow.
  2. Check that the expected Designer version is available.
  3. For the older Pika template, check its rgthree-comfy / Power LoRA Loader requirement.
  4. For historical Designer IDs, consult the matching archived release.

If the issue is damaged connection data rather than a missing node type, rgthree-comfy provides a Link Fixer at http://127.0.0.1:8188/rgthree/link_fixer when running ComfyUI locally on its usual port.

H3-Character-Sheet-Designer Frequently Asked Questions

Will every view preserve exactly the same character?
Not necessarily. Native image references help guide appearance, but do not guarantee exact identity, anatomy, garment fidelity, or geometry.

Can the extension accurately invent a character’s unseen back or hidden outfit details?
No. Unseen surfaces are generated interpretations. Additional relevant references can clarify your intent, but do not guarantee exact reconstruction.

Does each panel receive completely separate conditioning?
No. Reference roles and panel descriptions are text-based instructions. The Designer does not add independently encoded regional conditioning or hard BREAK separation.

Does a successful software test mean better artwork?
No. Interface tests, image pass-through checks, and workflow wiring checks are separate from GPU image-quality evaluation.

Why use a fixed seed?
It makes comparisons more informative by keeping the starting randomness consistent while you change a setting.

Can I compare two results more easily?
The optional rgthree-comfy Image Comparer supports slide and click comparisons. It is a separate utility, not a Designer feature.

Why does the older Pika workflow use multiple frames for a static sheet?
That template samples five frames and saves the first decoded frame. The current MiniMax H3 Image template instead generates a single static image.

Learn More about H3-Character-Sheet-Designer

H3-Character-Sheet-Designer License and Attribution

Keep the extension’s LICENSE, NOTICE.md, and artwork attribution.

The image models, text encoders, VAEs, and LoRA adapters have separate licenses and access conditions. Review those terms for the components you use.

H3 Character Sheet Designer Related Nodes

RunComfy
Copyright 2025 RunComfy. All Rights Reserved.

RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.

H3 Character Sheet Designer detailed guide | ComfyUI