---
title: "Minds for ComfyUI: image and video preference… | Minds"
canonical_url: "https://getminds.ai/guide/integration-comfyui"
last_updated: "2026-10-23T01:51:50.202Z"
meta:
  description: "Install Minds in ComfyUI, compare image and video creatives with selected Audiences, review the draft, and read preference results inside one node. Includes..."
  "og:description": "Install Minds in ComfyUI, compare image and video creatives with selected Audiences, review the draft, and read preference results inside one node. Includes..."
  "og:title": "Minds for ComfyUI: image and video preference… | Minds"
  "twitter:description": "Install Minds in ComfyUI, compare image and video creatives with selected Audiences, review the draft, and read preference results inside one node. Includes..."
  "twitter:title": "Minds for ComfyUI: image and video preference… | Minds"
---

Minds

October 16, 2026·Updated October 17, 2026·Minds Team # **Minds for ComfyUI: image and video preference testing** Minds adds one audience-preference node to ComfyUI. Connect images, videos with audio, hosted media or text concepts; choose existing Audiences; review the draft; then see results and pass the selected original media downstream. Minds brings audience research into the graph where you create images and videos. Instead of exporting every variation, opening a separate research setup and manually routing the preferred file back into production, you connect your candidates to _Minds · Audience Preferences_. The node asks your selected Audiences one reviewed preference question, displays the completed research, and returns the original selected media to the next nodes. [Minds](https://getminds.ai/guide/overview) is the end-to-end platform for commercial synthetic research. It connects qualitative and quantitative work through reusable Minds, Audiences and Studies. Minds PRISM provides the underlying reasoning, inference and source-modeling foundation for scoped synthetic research. The broader platform supports workflows such as qualitative exploration and quantitative methods including MaxDiff; this ComfyUI adapter provides a focused categorical preference test, rather than every research method in a graph widget. Explore [Minds and their grounding](https://getminds.ai/guide/minds) and the [broader feature guide](https://getminds.ai/guide/features) when your research needs go beyond choosing between creative alternatives.![A ComfyUI image comparison with two input images, one Minds node, an embedded preference chart, and the selected image saved downstream](https://getminds.ai/_ipx/_/images/videos/integrations/comfyui/coca-cola-workflow-transparent.png) This workflow-only capture replays a completed integration test using two published Coca-Cola campaign images. Ad B (Share a Coke) received 4 of 5 selections (80%); Ad A (football) received 1 (20%). Both original inputs, their matching result thumbnails and the saved winner are visible. This five-Mind QA run verifies the integration, not campaign performance. Campaign sources: [Coca-Cola football](https://www.coca-colacompany.com/media-center/coca-cola-invites-gen-z-to-live-fully-in-the-moment), [Share a Coke](https://www.coca-colacompany.com/media-center/iconic-share-a-coke-is-back-for-a-new-generation.html). Place the pointer over the results and scroll with the mouse wheel or a two-finger trackpad gesture to reach the findings and bottom controls. The graph stays in place. This taller documentation capture shows the entire result at once. ## What you can do in one node The node combines material collection, draft preparation, approved execution, result display and selection. Ordinary ComfyUI fields collect the question, viewing context, candidate labels and Audience IDs. The existing Minds response interface appears within the node after research completes. | Your material | How it enters the node | What can continue downstream |
| --- | --- | --- | | Generated images | Native IMAGE batches or lists on `images` and optionally `images_b` | Original selected images | | Generated videos | Native VIDEO objects or lists on `videos` and optionally `videos_b` | Original selected videos, including audio | | Hosted image or video files | Labelled HTTPS records in `assets_json` | Selected creative references | | Authored concepts or scripts | Labelled text records in `assets_json` | Selected concept records | | A mixed comparison | Supported inputs together, each with a distinct label | The selected native media that exists in the original inputs | The result remains tied to the material, exact question and selected populations. It does not silently choose a different Audience, replace missing media with a model description, or merge several Audiences into a single population. Read the [results and troubleshooting reference](https://getminds.ai/guide/comfyui-results-and-troubleshooting) for the precise output and retry behavior. ## Requirements Use a ComfyUI server you control, with permission to install a custom-node pack and its Python dependencies. The server needs outbound HTTPS access to Minds and a Minds account with API access and ready Audiences. Its Python environment must include ComfyUI's existing PyTorch installation. The pack adds its declared HTTP and image-processing dependencies to that environment. Prepare at least two distinct alternatives and a question that those alternatives can answer. An image comparison needs real images; a video comparison needs the intended videos if the question depends on motion, timing or audio. A text concept is useful for comparing proposed stories, but it is not an assessment of a finished video. Choose an Audience for the decision you are making. The [Audience guide](https://getminds.ai/guide/audiences) explains reusable populations, and the [Study guide](https://getminds.ai/guide/studies) explains the saved research object behind the node. A relevant-sounding Audience name alone does not establish grounding, representativeness or suitability for a consequential decision. ## Installation and current distribution The official package is [ComfyUI-Minds on GitHub](https://github.com/minds-ai-co/ComfyUI-Minds), published by Minds. The [Comfy Registry listing](https://registry.comfy.org/nodes/comfyui-minds) uses publisher `minds` and node-pack identifier `comfyui-minds`. Check the version's current verification status there before relying on Manager verification. The GitHub release provides a direct installation path independently of that status.![The public Comfy Registry listing with the official Minds logo, package name, installation command and description](https://getminds.ai/_ipx/_/images/videos/integrations/comfyui/registry-listing.png) As of October 16, 2026, Registry version `0.2.0` is flagged and awaiting review. The GitHub release is available, but the Registry submission does not yet establish verified Manager availability. Check the live version status before installation. The documented end-to-end evidence is for ComfyUI 0.41.0 with frontend 1.55.17 and short illustrative image/video fixtures. It establishes that installation, approved research, result rendering and native saving worked in that environment. It does not establish Comfy Cloud execution, marketplace Partner Node acceptance, or compatibility with every future ComfyUI version. ### Install from the official repository Open a terminal in your ComfyUI installation and clone the pack into `custom_nodes`:```
cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/minds-ai-co/ComfyUI-Minds
``` Install `ComfyUI-Minds/requirements.txt` with the _same Python interpreter that runs ComfyUI_. For example, replacing the interpreter path with your actual environment:```
/path/to/comfy-python -m pip install -r ComfyUI-Minds/requirements.txt
``` A global `pip` command may install dependencies into a different environment. Desktop, portable and virtual-environment installations have different interpreter locations; use the location belonging to the server you actually launch. Do not replace ComfyUI's PyTorch installation simply to install this pack. Restart ComfyUI after installation. Search the node menu for _Minds · Audience Preferences_. If it does not appear, inspect the server's custom-node import error before attempting a research run. Installing a repository in the wrong folder or interpreter will not be fixed by changing your Minds question. ### Provide the server credential Create or manage your key in [Minds API-key settings](https://getminds.ai/?settings=api-keys), then configure your credential manager to inject it as `MINDS_API_KEY` into the ComfyUI server process. Restart or relaunch that process with the configured environment. A variable set in an unrelated terminal does not automatically reach an already running desktop application or service. The key is used by the Python adapter to call Minds. It is not a candidate input, a property to embed in a shared graph, or a value required by the response iframe. Keep the workflow reusable by keeping authentication separate from its question and media. ## Load a runnable example The pack includes seven example workflows covering ad selection, thumbnails, packaging, Audience differences, original-versus-revision comparisons, video concepts and native video selection. Open one through ComfyUI's template browser or import its JSON from `example_workflows` in the installed package. For the included illustrative media, copy the example assets into ComfyUI's `input` directory first. For your own test, choose your own files or connect your generation pipeline. Each example contains one Minds research node; the Load and Save nodes belong to the surrounding ComfyUI workflow. Start with a small, relevant comparison so that labels, attachments and the reviewed question are easy to inspect. Then expand the candidate set within the supported limits. Keep your first integration check separate from the creative decision you intend to make with a larger, appropriately grounded research population. ## Choose the Audiences Enter one or more existing Audience UUIDs in `audience_ids`, separated by commas. They must be visible to the account that owns the server's API key and ready for research. The node does not interpret a demographic sentence in this field or create a new Audience from it. An optional _Minds · Find Audiences_ node lists accessible IDs, names, Mind counts and readiness. Use its search field to narrow the list; use its offset to page beyond the first hundred entries. This helper is for discovery. Your research graph still needs only the single preference node once you have selected the IDs. For a comparison across Audiences, retain separate results. If you also want the graph to select media automatically, choose one of the tested IDs in `selection_audience_id`. Leaving that field blank for a multi-Audience test displays the research but blocks automatic selection, because the adapter has no basis for deciding which population should determine production. ## Connect images and label them exactly Connect an IMAGE batch to `images`. A second source can connect to `images_b`, avoiding an extra Minds collection node when comparing two native sources. The node supports lists as well as batches; image dimensions may differ when outputs continue as separate images.`labels_json` names every native candidate in order. Images come first, with the first image input before the second; native videos follow, with the first video input before the second. Labels are exact, distinct identities used by the question, result and selector.```
["Original", "Revision"]
``` If a connected batch contains more than one image, each image needs a label. Two connected batches can therefore contain more than two candidates. Changing the candidate count without updating the labels is a contract error, not permission to invent names or drop an image. Choose labels that remain understandable later, such as “Original headline” and “Short headline.” Avoid putting an expected winner in the label. The research asks about the supplied alternatives; labels should help identify them rather than add a persuasive claim of their own. ## Write one useful preference question The node runs a fixed categorical question over the labelled alternatives. Ask for one judgment that the material can support. “Which thumbnail would you prefer to open?” is different from “Which thumbnail explains the subject more clearly?” A result from one question does not answer the other automatically. Use `context` for information the Audience needs to interpret the task: where the creative will appear, the intended viewing situation, or what should be considered. Keep the context balanced across alternatives. If the product, offer or audience relationship differs, decide whether you are testing a creative treatment or a broader proposition before interpreting the result. The complete question, context and creative references must fit the API's fixed-question limit. Do not solve an overflow by silently removing the part that defines the viewing situation. Rewrite the brief deliberately and start a new test request for the changed content. ## Queue, review and approve Queue the graph once. The node creates or reuses its saved Study, uploads native media as needed, and prepares the exact draft. A review dialog displays the selected Audiences, language, complete question and alternatives. Preparing the draft does not start respondent research; downstream selection remains blocked.![The Minds draft-review dialog before research execution, showing the question, Audience and candidate alternatives](https://getminds.ai/_ipx/_/images/videos/integrations/comfyui/draft-review.png) Read the question as the respondents will receive it, including its context and candidate references. Confirm that the label order matches the connected material and that the selected population fits the decision. Cancel if something is wrong. Approval applies to the displayed saved draft, rather than to any later change to the graph. Choose _Approve this draft_, then queue again. The node verifies the approved revision and starts or resumes that draft's durable research run. Account access, Audience readiness, plan and usage limits still apply through the shared Minds API. Use _New test request_ after changing material, labels, question, context, language or Audiences. That button creates a fresh request identity and clears the old approval and response view. Keep the existing request identity when resuming an interrupted attempt with the same inputs. ## Read the embedded results Completed research appears inside the node through the same Minds response surface used by MCP result widgets. It can show the question, Audience breakdowns, categorical distributions, findings and available individual Mind responses. Scroll within the result area to inspect more of the response.![The actual Minds response widget showing the illustrative image preference distribution](https://getminds.ai/_ipx/_/images/videos/integrations/comfyui/coca-cola-response.png) The screenshots use one five-Mind QA Audience and intentionally simple fixtures. They demonstrate a functioning integration path. The Audience's display name and a visually clear bar chart do not convert those fixtures into representative market evidence or a calibrated campaign forecast. For a substantive decision, inspect whether the Audience, viewing context and alternatives match the task. Examine the finding and available answers alongside the distribution. Open the Study in Minds when you need the broader saved research workflow, comparison or export facilities described in the [Study documentation](https://getminds.ai/guide/studies). ## Continue with the selected originals Connect `selected_images` to an image-processing or Save Image node. Connect `selected_videos` to video-processing or Save Video. These outputs retain the original native objects, rather than reconstructing a winning image from a thumbnail or turning a video into sampled stills. Set `top_k` to choose a shortlist size within the candidate count. Ties at the cutoff remain included, so the output can contain more items than that value. A tied first place is a tied result; the node does not select the first input merely because it appeared first. Hosted assets and text concepts can be selected as creative records, but cannot create a native IMAGE or VIDEO object that was never connected. An absent native modality blocks that media output while the other completed result outputs remain usable. The [results reference](https://getminds.ai/guide/comfyui-results-and-troubleshooting) explains the JSON and custom outputs for automation. ## Test videos and concepts deliberately For native videos, connect the intended VIDEO objects, label them and ask a question grounded in the complete clip. The audio track matters when the question depends on narration, music, rhythm or spoken information. The [video workflow guide](https://getminds.ai/guide/comfyui-video-preference-testing) covers the ports, review and native output in detail. For proposed stories or scripts, use labelled text concepts in `assets_json`. State that the task concerns concepts, not completed videos. A preference for one authored concept does not establish preference for an eventual execution with unknown casting, edit, sound, pace or visuals. The [illustrated creative-workflow article](https://getminds.ai/blog/comfyui-audience-preference-testing) walks through practical uses, including ad variants and original-versus-revision comparisons. Keep those examples tied to the question you actually run rather than treating every preference distribution as the same commercial outcome. ## How this differs from the ChatGPT integration The [ChatGPT integration](https://getminds.ai/guide/integration-chatgpt) brings Minds research into a conversation through MCP and OAuth. ComfyUI's adapter uses the Minds API from its Python server and a server-side key. Both reach the shared research platform, but they expose different host workflows and authentication paths. Use ComfyUI when native generated media should remain in the graph and a preference result should route original files downstream. Use a conversational client such as [ChatGPT](https://getminds.ai/guide/integration-chatgpt) or [Claude](https://getminds.ai/guide/integration-claude) when the surrounding task is a research dialogue or a broader agent workflow. The [integration overview](https://getminds.ai/guide/integrations) explains the available host options. The ComfyUI results iframe reuses the response interface; it does not make the node an MCP server, add OAuth to ComfyUI, or require you to install a ChatGPT plugin. Likewise, the node does not install a Mosaic checkpoint or provide a model-selection setting. Its supported behavior is the tested Minds Study API workflow described here. ## Next steps and support Follow the [video guide](https://getminds.ai/guide/comfyui-video-preference-testing) for a native audio/video comparison, or the [results and troubleshooting reference](https://getminds.ai/guide/comfyui-results-and-troubleshooting) when an output is blocked, a request times out or a result needs interpretation. The [source repository](https://github.com/minds-ai-co/ComfyUI-Minds) contains the installable pack, examples and support issue tracker. If you are new to Minds, start with [getting started](https://getminds.ai/guide/overview), [Minds](https://getminds.ai/guide/minds), [Audiences](https://getminds.ai/guide/audiences) and [Studies](https://getminds.ai/guide/studies). The graph is a focused entry point into that connected research workflow, and the quality of the decision still depends on the population, material and research question you bring to it. ## **Frequently asked questions**### **Is Minds for ComfyUI one node?** Yes. Minds · Audience Preferences accepts the material and research settings, prepares a reviewable draft, runs it after approval, displays the response widget, and returns ranking and selected-media outputs. An optional Find Audiences helper supplies account-visible Audience IDs. ### **Does the node support videos with audio?** Yes. Native ComfyUI VIDEO objects are uploaded as MP4 with audio. The winning original VIDEO objects continue downstream with their audio. The published integration has end-to-end evidence for short video fixtures; that does not establish acceptance of every long-video format or duration. ### **Can I choose my own Audience?** Choose existing, ready Minds Audiences that the server's API-key account can access. The node returns separate preference results for each selected Audience. It does not create a population from an ungrounded demographic label typed into a field. ### **Where does the Minds API key go?** Provide MINDS_API_KEY to the ComfyUI server process through your credential manager. It belongs in neither a normal input widget nor workflow JSON. The embedded results widget receives answer data without receiving the key. ### **Does the node predict actual campaign performance?** It estimates synthetic preference among the supplied alternatives. It does not measure clicks, conversions, sales or representative human population shares. Use the result to decide what to investigate or validate next. ### **Is the integration available on Comfy Cloud?** The documented and tested installation is a custom-node pack on your own ComfyUI server. GitHub distribution and Comfy Registry submission do not establish Comfy Cloud execution or native Partner Node availability. [Minds](https://getminds.ai/)© 2026 Minds. Your target audience. AI-driven and grounded in transparent evidence. Build within minutes. [Minds on X (Twitter)](https://x.com/mindsai_co) [Minds on LinkedIn](https://www.linkedin.com/company/mindsaicompany/) [Minds on Instagram](https://www.instagram.com/getminds.ai/)Minds is part of [![ESOMAR](https://getminds.ai/images/newsroom/logos/esomar-logo.svg)ESOMAR](https://esomar.org/) [![GreenBook](https://getminds.ai/images/newsroom/logos/greenbook.svg)GreenBook Directory](https://greenbook.org/company/Minds) [![Insight Platforms](https://getminds.ai/images/newsroom/logos/insight-platforms.png)Insight Platforms](https://www.insightplatforms.com/platforms/minds/) [![Capterra](https://getminds.ai/images/newsroom/logos/capterra.svg)Capterra](https://www.capterra.com/p/10046203/Minds/) [![G2](https://getminds.ai/images/newsroom/logos/g2.svg)G2](https://www.g2.com/products/minds/reviews) [![CSSDA Best UX Design Award](https://getminds.ai/images/newsroom/logos/cssda-best-ux-award.png)CSSDA Best UX Design Award](https://www.cssdesignawards.com/) [![CSSDA Best Innovation Award](https://getminds.ai/images/newsroom/logos/cssda-best-innovation-award.png)CSSDA Best Innovation Award](https://www.cssdesignawards.com/) [![CSSDA Best UI Design Award](https://getminds.ai/images/newsroom/logos/cssda-best-ui-award.png)CSSDA Best UI Design Award](https://www.cssdesignawards.com/)