ComfyUI-KIE-Nodes-Next Introduction
ComfyUI-KIE-Nodes-Next brings KIE.ai models into ComfyUI as individual nodes. Instead of filling out a general-purpose API form, you can choose a node for the image, video, audio, or text model you want to use and connect it to the rest of your graph.
The extension also includes local tools for planning shots, reviewing references, comparing results, and assembling a finished video. These tools can help you organize an AI art project before you spend credits on generation, then keep track of the choices and results you want to reuse.
The current documented release is v0.6.3. It improves model-specific inputs and reconciles the extension’s catalog with KIE’s Market listings. It does not complete every item in the project’s roadmap, and paid workflows still need to be checked against the selected model’s requirements.
Useful project references include the release notes, the model-controls review, and the starter workflows.
How ComfyUI-KIE-Nodes-Next Works
Think of a ComfyUI graph as a production desk: one node prepares an idea or reference, another sends a generation request, and later nodes help you inspect or save the result. ComfyUI-KIE-Nodes-Next makes KIE’s documented models available as nodes on that desk.
- Choose a model node. Nodes are grouped by media type, provider, and model family. Each exposes inputs drawn from that model’s documented API, such as a prompt, duration, aspect ratio, or reference image.
- Connect media directly. Where a model supports it, you can connect ComfyUI
IMAGE,VIDEO, orAUDIOoutputs. The extension handles the upload and passes the resulting media URLs to KIE. - Check and run. Local checks catch some missing or incompatible inputs before upload. Running a generation node then submits a task to KIE; a local planning or review node does not submit a generation unless you connect and run a paid model node.
- Use the outputs. Generation nodes return native ComfyUI media and task information. Depending on what KIE reports, they can also provide result URLs, raw response data, and credit usage.
For example, you might draft a six-shot sequence, inspect it as a contact sheet, adjust the prompts, and then connect the approved plan to a compatible video model. The contact sheet is a planning aid; the model node is the part that requests generated media.
The model library is built from KIE’s documentation and Market information, with cached definitions available when discovery is temporarily unavailable. Because KIE changes its catalog, the nodes and accepted controls can change over time. A field appearing in a node does not guarantee that every combination of settings will be accepted by the provider.
ComfyUI-KIE-Nodes-Next Features
Model-specific controls and input checks
Individual model nodes expose the fields documented for their operations. Depending on the model, these may include prompts, image or video references, duration, resolution, audio options, or text-model controls.
Right-click a generation or Studio node and choose KIE Next · Check inputs & model controls to inspect its cached documented fields, limits, and camera-control support. This inspection does not upload media or start a generation. Connected values are checked when the workflow runs.
Local checks can identify issues such as missing required inputs, invalid JSON, or unsupported numeric values before an upload. They cannot verify every provider rule or judge whether a reference image is suitable. If a capability is unknown, the extension labels it as unknown rather than assuming support.
Storyboards and shot planning
For a quick visual plan, connect KIE • Shot Sequence to KIE • Storyboard Contact Sheet. You can add up to six hero-frame images from Load Image or another image node. The contact-sheet node saves a persistent image in ComfyUI’s output folder and returns both an IMAGE and JSON notes. It runs locally.
For a written script, use:
KIE • Script to Shot Draft → KIE • Shot Plan Review → KIE • Storyboard Contact Sheet
The draft can split text on blank paragraphs or treat each line as a shot. You can set a target duration and choose an optional visual rhythm, such as cinematic reveal, documentary observer, product detail, action build, or interview coverage. The neutral default does not impose a style. These options offer editable camera suggestions; the planner does not interpret story meaning or invent new shots.
You can inspect the draft’s JSON, edit prompts, timing, and camera suggestions in Shot Plan Review, then pass the reviewed sequence to Storyboard or Kling Omni Studio. For Kling Omni manual multi-shot work, keep the total within its documented 3–15-second duration; the extension checks this before upload.
Character and production references
Character Packs keep approved identity material together. KIE • Save Character Pack stores a portrait, optional additional views or expressions, and identity and continuity notes. Load Character Pack returns an identity prompt, while Character Reference supplies an approved image to an operation that documents reference-image support. Character Consistency Test Plan prepares eight test cases for you to run and judge; it does not generate them automatically.
A Character Variant adds an approved outfit or appearance to a specific, immutable Character Pack version. This keeps a wardrobe change separate from the character’s core identity. A reference and prompt can improve continuity, but they cannot guarantee that every model will reproduce a person consistently. Saving a pack marked as a real person’s likeness requires confirmation of permission.
Save Element, Load Element, and Element Reference serve a similar purpose for locations, props, and styles. Locations and props require an approved image; a style may be text only. Changes to saved content require a new version. Consistency Board places an approved reference beside up to four candidate images for human review without claiming to calculate an identity score.
Camera, variations, and visual comparison
KIE • Camera Path provides a draggable top-view path with dolly, orbit, crane, and custom presets, plus a saved preview image. Its direction becomes prompt guidance unless the chosen model explicitly documents a matching camera control.
KIE • Variation Matrix prepares four prompts, each with one deliberate change. You decide which takes to connect to paid generation nodes. After generating, KIE • Variation Board places up to four results and their intended changes on a persistent 2×2 sheet so you can compare them visually.
KIE • Camera Director builds reusable wording for shot size, angle, movement, lens, speed, stabilization, and focus. KIE • Shot Sequence plans up to six shots with validated durations.
Video review and sound planning
KIE • Video Event Analysis samples a connected video locally and identifies timecoded visual-change candidates. It does not recognize actions, speech, or sounds. KIE • Music Brief from Video turns those markers or your own event descriptions into an instrumental brief. KIE • Sound Cue Sheet uses editor-described sound events for effects and keeps unverified visual changes separate. These planning steps do not generate audio.
For closer inspection, KIE • Sample Video Frames produces labelled frames, timestamps, and a contact sheet. You can intentionally send selected frames and the suggested prompt to a vision-capable model, but that is a separate, potentially paid step. A model may miss events between frames or describe something incorrectly. Watch the original video, correct the returned JSON, and paste it into KIE • Review Video Events. Only descriptions marked editor_reviewed flow onward as verified events.
KIE • SFX Prompt from Cue prepares an approved event or ambience prompt of up to 500 characters, along with cue timing and other options, for a compatible KIE Suno Sounds (https://docs.kie.ai/suno-api/generate-sounds) node. Connecting and running that generation node is an explicit paid action.
Video finishing and reuse
KIE • Assemble Video + Music can join up to four compatible video takes with hard cuts, fit one music track, and mix explicitly connected dialogue and up to two timed sound effects. Short music can be padded with silence or looped with a crossfade. Optional music ducking responds to audio level, not speech recognition. Source-video audio is replaced, so connect any dialogue you need separately and listen to the exported result to check timing and levels. The node does not yet offer more than two effects or alternative transition styles.
KIE • Preview Video plays a connected VIDEO inside a node. KIE • Save Video exports it; both pass the video through and provide a saved_file value that KIE • Persistent Load Video can use in a later workflow. Previews are kept in output/KIE-Previews/, and saved exports use output/KIE-Videos/ by default.
KIE • Reuse Completed Image/Video/Audio retrieves an already successful KIE task by ID without submitting it again. Save Generation Recipe records a versioned snapshot of the model, exact prompt, references, camera plan, task ID, result URLs, and confirmed spend; Load Generation Recipe retrieves it. Saving a recipe is currently an explicit step, not an automatic action after every generation.
Tasks, credits, and workflow shortcuts
KIE • Task Board displays the status, progress, result URLs, and provider-reported spend for the last 20 locally submitted tasks, or for up to 20 task IDs you enter. Refreshing the board does not resubmit them. Its local journal records task IDs and status information, not prompts or API keys; it is not an account-wide history.
After successful generation, model and Studio nodes display reported credits spent and the live credits left balance when available. If task-level spend is unavailable, a balance difference is labelled as an estimate and is not added to confirmed local spend. The Connection → Credits settings show the latest balance and cumulative provider-reported spend tracked by this ComfyUI profile. Refresh balance now checks the balance without generating media.
You can assign a KIE Next node to one of nine favorite slots from its right-click menu, then bind KIE Next: Insert favorite node 1–9 in ComfyUI’s keybindings. Favorites belong to the current frontend profile and do not replace existing ComfyUI or browser shortcuts.
Four examples in templates/ cover video-event review, music and sound briefing, finishing video with music, and comparing character consistency. As supplied, none calls a paid model. Import one into ComfyUI, choose your own files, and review its prompts and cues before using it in a production graph.
ComfyUI-KIE-Nodes-Next Models
The extension’s catalog is broader than a fixed list, so the examples below are not every available node. Nodes are organized under KIE Next by media type, provider, and family:
| Area | Documented examples | Good starting point when you want to… |
|---|---|---|
| Video | ByteDance Seedance, Kling, Wan, Runway, Google Veo | Generate or edit moving images using an operation whose inputs match your source material. |
| Image | ByteDance Seedream, Google Nano Banana, OpenAI GPT Image, Black Forest Labs Flux | Create or transform still images using the selected model’s documented controls. |
| Audio | Suno, ElevenLabs, Google Gemini TTS | Work with music, sounds, or speech through a compatible operation. |
| Text / LLM | Claude, OpenAI GPT and Codex, Google Gemini, xAI Grok | Generate text for prompts, briefs, or other graph steps. |
The first-launch bootstrap includes definitions for Seedance 1.5 Pro, Seedance 2.0, Seedance 2.0 Fast, Seedance 2.0 Mini, Seedance 2.5, Claude Opus 5, GPT 5.6 Sol, Suno Generate Music, and Veo 3.1 Generate Video. Catalog discovery can add more nodes and documented versions. A model name alone does not tell you its price, speed, or supported input modes; inspect that node’s controls before choosing it.
Some families have dedicated Studio nodes:
- Kling 3.0 Omni Studio supports documented single-shot and multi-shot approaches, image-guided generation, audio, and 4K controls. For manual multi-shot sequences, check the total duration before running.
- Seedance Studio presents distinct text, first-frame, first-and-last-frame, and multimodal-reference modes. Seedance 2.0, Fast, and Mini also expose documented controls such as reference media, generated audio, resolution, aspect ratio, and duration. First-frame modes and multimodal-reference mode are mutually exclusive scenarios, so choose the one that matches your intended workflow.
- Wan 2.7 - Video Edit edits a selected source clip as a whole. Trim the clip with ComfyUI’s
VideoTrimfirst if needed, then provide a documented 2–10-second source and an optional nativeIMAGEreference. This is a paid provider edit, not a spatial mask editor. See the Wan 2.7 Video Edit API documentation (https://docs.kie.ai/market/wan/2-7-videoedit) for accepted fields and limits.
Text-model nodes return normal STRING text along with response, usage, and credit information. Depending on the operation, controls may include a system prompt, history, tools, token limit, reasoning setting, streaming, web search, or an image input. Use only the controls shown for your selected node.
The extension’s October 22 catalog check found 230 live Market entries; 229 supplied usable OpenAPI documents. The combined inventory contained 250 IDs, including legacy IDs and provider versions, and 266 operations. These are different ways of counting the catalog, not 250 newly released models. Gemini 3.8 Flash TTS returned a null schema in that check, so its last documented definition was retained.
What’s New with ComfyUI-KIE-Nodes-Next
| Version | Documented change | Why it matters |
|---|---|---|
| v0.3 | Introduced individual, model-specific nodes in place of the user-facing Any KIE API node; added native media connections and larger prompt editing areas. | You can build ordinary workflows around the model you actually intend to use, with less manual URL handling. |
| v0.5.0 | Added reusable, versioned Character Packs and a manual consistency test plan. | You can preserve approved references and test a character across different views without silently changing the original pack. |
| v0.6.0 | Added Preview Video and Save Video nodes. | You can inspect a VIDEO in the graph and export it for later use. |
| v0.6.3 | Reconciled live Market listings with documentation, restored missing IDs and alternative input modes, and improved schema-based controls while preserving existing workflow node IDs. | More documented model options and limits appear in the appropriate nodes without intentionally replacing IDs used by existing graphs. |
Other documented capabilities include Studio direction tools, storyboard review, production-reference libraries, local video and sound planning, task and credit displays, and generation recipes. The supplied context does not assign each of these features to a particular release, so they are not presented as version-specific changes here. Consult the release notes for the project’s release history.
Troubleshooting ComfyUI-KIE-Nodes-Next
A model rejects your inputs
- Right-click the node and select KIE Next · Check inputs & model controls.
- Check required fields, allowed values, numeric limits, and the selected input variant.
- If media is connected, confirm that it matches the operation’s documented type and limits. Local checks cannot inspect every content restriction.
- Run again only after correcting the input. If KIE still rejects it, consult that model’s documentation for conditions the local checks cannot cover.
For Seedance workflows, do not combine first-frame or first-and-last-frame inputs with a mutually exclusive multimodal-reference scenario. For Kling Omni manual multi-shot work, check the total duration. For Wan 2.7 - Video Edit, check source duration and the documented media requirements before a paid submission.
A node or model is missing
- Look under the relevant KIE Next media category, provider, and model family.
- Allow the extension’s catalog discovery to complete and the ComfyUI frontend to reload if prompted.
- Check whether KIE documents the operation as a callable model/API page. The extension cannot promise a node for an undocumented capability.
- If discovery is temporarily unavailable, the cached library remains usable; check again when KIE’s documentation and Market services are available.
For catalog or control discrepancies, the model-controls review provides project-specific context.
Your API key or balance is not shown as expected
Open ComfyUI → Settings → KIE.ai Nodes Next → Connection. A saved key is represented by a masked fingerprint and connection status, not its full text. You can test the saved key or replace it there; a replacement is validated before it overwrites the existing key. The key is stored outside workflow JSON, and environment-variable overrides are not supported.
Use Refresh balance now to query the balance without starting a generation. If the balance service is unavailable after a successful task, the media result can still be returned while the balance is shown as unavailable. Local tracked spend starts with this feature and is not KIE’s account-wide billing history.
A reused character looks different
- Confirm that you loaded the intended Character Pack name and version.
- Check whether the selected operation documents reference-image support; a connected prompt alone is not an identity lock.
- Keep identity notes separate from scene, outfit, and lighting changes.
- Compare candidates against the approved image with Consistency Board, then approve a new variant or version deliberately.
A video brief describes something that did not happen
Sampling finds frames and visual changes, not a complete understanding of the clip. Watch the source video, correct any model-generated event JSON, and mark it editor_reviewed in Review Video Events before treating those descriptions as verified. Do not turn visual-change markers into sound effects unless you have described the sound event yourself.
The final video has missing dialogue or an uneven mix
Assemble Video + Music replaces source-video audio. Connect required dialogue explicitly, check sound-effect cue times, and listen to the saved export. Music ducking responds to audio level rather than detecting speech, so adjust its amount by ear.
Frequently asked questions
Do planning and review nodes spend KIE generation credits?
The described storyboard, comparison, video-analysis, and briefing helpers run locally. A connected generation node is a separate paid action when you run it.
Does checking a node’s inputs upload my images or start a task?
No. The right-click input and controls check inspects cached definitions. Connected values are checked during workflow execution.
Can I recover a completed result without generating it again?
Yes. Use the appropriate Reuse Completed Image/Video/Audio node with a successful KIE task ID.
Are credit estimates the same as confirmed spend?
No. When KIE does not report task-level spend, the extension labels a balance-based difference as an estimate; concurrent jobs can affect it.
Does a reference image guarantee character consistency?
No. Results depend on the selected model, its supported reference mode, and the content of your prompt and images. Review outputs visually.
Learn More about ComfyUI-KIE-Nodes-Next
- Starter workflows show local review, audio-planning, video-finishing, and character-comparison graphs. Replace the example file selections and review each prompt before use.
- Release notes provide the extension’s release history.
- Model-controls review documents the controls audit and planned improvements.
- Roadmap distinguishes planned work from completed features.
- KIE Suno Sounds documentation (https://docs.kie.ai/suno-api/generate-sounds) explains the compatible sound-generation operation used with an approved SFX prompt.
- Wan 2.7 Video Edit documentation (https://docs.kie.ai/market/wan/2-7-videoedit) lists that operation’s accepted inputs and limits.
The project is licensed under the MIT license.
