Run image, video, speech, music and transcription models from deAPI inside ComfyUI. Jobs run on deAPI's cloud GPUs, so you don't need a local GPU or model downloads. You pay per job at deAPI's prices.

All nodes use standard ComfyUI types (IMAGE, MASK, VIDEO, AUDIO, STRING), so they connect to core nodes such as Load Image, Save Image, Save Video and Save Audio, and to any other custom nodes.

Install

ComfyUI Manager / Registry (recommended)

Search for deAPI in ComfyUI Manager, or run:

comfy node install comfyui-deapi

Manual

cd ComfyUI/custom_nodes
git clone https://github.com/deapi-ai/comfyui-deapi
pip install -r comfyui-deapi/requirements.txt   # only needs `requests`, which ComfyUI already ships with

Restart ComfyUI. The nodes appear under the deAPI category, and you can find them by searching "deAPI".

API key

Create a free account at https://app.deapi.ai/register and copy your API key from the dashboard.
Pick one of these:
- Environment variable (recommended): set DEAPI_API_KEY before starting ComfyUI.
  - Windows (PowerShell): $env:DEAPI_API_KEY="your-key"; python main.py
  - macOS / Linux: DEAPI_API_KEY=your-key python main.py
- Node input: paste the key into the optional api_key field on a node. Keys typed there are saved inside the workflow, and ComfyUI also embeds the workflow in the metadata of every PNG it saves. Don't share those workflows or images; use the environment variable instead.

Nodes

| Node | Inputs | Outputs | Models | |---|---|---|---| | deAPI Text to Image | prompt, model, width, height, steps, seed, negative_prompt | IMAGE | Z-Image-Turbo, FLUX.2 Klein 4B, FLUX.1 Schnell, Z-Anime Distill | | deAPI Image Edit | image, prompt, model, steps, seed, image_2, image_3, negative_prompt, width, height | IMAGE | FLUX.2 Klein 4B (fast, up to 3 reference images), Qwen Image Edit Plus (slower, several minutes) | | deAPI Remove Background | image | IMAGE, MASK | Ben2 | | deAPI Upscale Image | image | IMAGE (4x) | RealESRGAN x4 | | deAPI Text to Video | prompt, model, width, height, frames, seed, negative_prompt | VIDEO, IMAGE (frames), STRING (file path) | LTX-2.5 22B, LTX-2.3 22B, MiniMax H3 33B, LTX-Video 13B | | deAPI Image to Video | first_frame, prompt, model, width, height, frames, seed, last_frame, negative_prompt | VIDEO, IMAGE (frames), STRING (file path) | same as Text to Video | | deAPI Text to Speech | text, voice, speed | AUDIO | Kokoro (41 voices in US/UK English, Spanish, French, Hindi, Italian and Brazilian Portuguese) | | deAPI Text to Music | caption, lyrics, model, duration, seed, bpm | AUDIO | ACE-Step 1.5 XL Turbo, ACE-Step 1.5 Turbo | | deAPI Transcribe Audio | model, include_timestamps, audio, source_url, language | STRING | Whisper Large V3 | | deAPI Image to Text (OCR) | image | STRING | Nanonets OCR |

Inputs in italics are optional. Every node also has an optional api_key input.

How the nodes behave:

Seed: -1 lets deAPI pick a random seed, and the node runs again on every queue. Any other value is reproducible.
Limits: each model has its own size, step and frame limits. If a value is out of range, the node clamps it to the nearest allowed value and logs a warning in the console. Video fps is fixed per model: 24 fps for LTX-2.x and MiniMax, 30 fps for LTX-Video 13B.
Video outputs: video nodes return a native VIDEO, so you can connect Save Video or Preview Video. They also return the decoded frames as IMAGE and the path of the MP4, which is saved in ComfyUI/output/deapi/.
Remove Background mask: the MASK output is 1 on the subject. To get a transparent PNG, connect it through InvertMask into Join Image with Alpha, then Save Image. The example workflow shows this.
Transcribe Audio: accepts an AUDIO input or a source_url (YouTube, X, Twitch, Kick, or a direct media link). The text appears on the node and is also returned as a STRING.
Batches: image nodes process every image in an input batch.
Progress and cancelling: progress shows on the node, and status lines are printed to the console. Cancel stops waiting for a job, but a job that was already submitted may still finish (and be billed) on deAPI.

Example workflows

The example_workflows/ folder has ready-made workflows. Drag one onto the ComfyUI canvas to load it, or open it from the Templates browser under Custom Nodes.

deapi_text_to_image.json: text to image, then Save Image.
deapi_image_edit_and_remove_background.json: edit an image, cut out the subject as a transparent PNG, and upscale it 4x.
deapi_image_to_video.json: generate a still, animate it, then Save Video.
deapi_text_to_video.json: text to video, then Save Video.
deapi_speech_music_transcribe.json: text to speech (saved and transcribed back to text) and text to music.

Typical timings

These are wall-clock times measured from ComfyUI, and they vary with load.

| Task | Time | |---|---| | Text to image (768x768) | 10-20 s | | Image edit, FLUX.2 Klein | 5-25 s | | Remove background | 5-10 s | | Upscale 4x | 20-50 s | | Text to speech, transcription, OCR | 5-15 s | | Image edit, Qwen Image Edit Plus | several minutes | | Text or image to video | 1-3 min |

Errors

Errors appear in the ComfyUI error dialog, and they include the HTTP status and deAPI's own message. These are the common ones:

deAPI HTTP 401 [invalid_key]: the key is missing or wrong. Check DEAPI_API_KEY or the node's api_key field.
deAPI HTTP 422: ... balance ...: your balance is too low. Top up at https://app.deapi.ai.
deAPI HTTP 422: The <field> field is ...: a parameter was rejected. The message names the field.
deAPI HTTP 429: you hit the rate limit. The node retries automatically a few times before giving up.
timed out after N s: the job took longer than the node waits. The message includes the request id, so you can look the job up later.

Configuration

| Environment variable | Purpose | |---|---| | DEAPI_API_KEY | API key used when a node's api_key field is empty | | DEAPI_BASE_URL | Override the API base URL (default https://api.deapi.ai/api/v2) |

comfyui-deapi

Nodes (10)

deAPI nodes for ComfyUI