DMAI NODES
A clear image workflow for ComfyUI. Slate surfaces, readable Inter typography and English controls.
0.2.0 preview · GitHub releases · Hugging Face download mirror
Prompter → Generation Engine → Gallery, with an ordered LoRA Loader feeding the engine.
What is included
| Node | Purpose |
|---|---|
| DMAI Prompter | Prompt, 19 aspect ratios, 1K / 2K / Custom dimensions, image count, Generate button and generation progress outline. |
| DMAI LoRA Loader | Roomy two-row cards with full filenames, enable controls, independent Model / CLIP strengths, reordering and strict file checks. |
| DMAI Generation Engine | Installed checkpoint and diffusion model picker, Manual sampling by default, uploaded DMAI Enhanced settings, seed sequence and execution report. |
| DMAI Gallery | Persistent PNG originals, multiple selection, per-image download and selection ZIP. |
| DMAI Krea Enhancer | A standalone Krea-only model node, using the optional upstream Krea2T Enhancer. |
| DMAI VAE Encode / Decode | Standard Comfy VAE operations, usable separately and with the engine's advanced inputs. |
Sampler and scheduler choices come from the running ComfyUI installation. Optional SAMPLER and SIGMAS connections support custom sampling workflows, including RES4LYF sampler nodes. See RES4LYF integration for the correct outputs and current limits.
The simple path loads models inside the engine. Advanced workflows can supply MODEL, CLIP and VAE together, and an encoded LATENT. Krea enhancement is only available for Krea profiles. A settings preset never installs a new model adapter.
The larger Prompter puts the prompt, image format, count and Generate control together. Engine exposes its sampling settings in Manual; switch to DMAI Enhanced to upload a settings JSON. Readable captions sit outside the native Canvas connection dots, and minimum node sizes keep controls and ports usable when resizing.
Install
From your ComfyUI directory, using the same Python environment as ComfyUI:
git clone https://github.com/ceroomedia/DMAI-NODES.git custom_nodes/DMAI-NODES
python -m pip install -r custom_nodes/DMAI-NODES/requirements.txt
Restart ComfyUI and refresh the browser. Search for DMAI in the node library.
Using an older DMAI package too? Since version 0.1.3, this package's Prompter has its own internal ID, DMAINodesPrompter. Choose DMAI Prompter from the DMAI-NODES package in category DMAI NODES. The older Suite and standalone Prompter keep their existing IDs. See updating from 0.1.2 for saved workflows and API clients.
- Windows Portable, local environments and Runpod
- Model files, presets and enhancement
- API workflow and future website integration
- Interface design
- Validation results and limits
ZIP installation is also supported: extract the release so ComfyUI/custom_nodes/DMAI-NODES/__init__.py exists. A Hugging Face mirror includes the package and checksum; this repository contains code and documentation, no model weights.
First workflow
- Import a starter from
workflows: Krea, SDXL, Qwen or FLUX. Use the.jsonfile for the UI and.api.jsonfor API calls. - Select an installed file at the top of Engine under Checkpoints or Diffusion models. Open Model files to set its Architecture and any separate text encoders and VAE. Checkpoint loading currently supports SDXL; diffusion loading supports Krea 2 Turbo, Qwen Image and FLUX.1 Dev.
- Keep Manual to adjust Steps, CFG, Sampler, Scheduler and Denoise. For a prepared settings file, switch to DMAI Enhanced and choose Upload JSON. Enhanced requires a matching preset before generation; returning to Manual restores your manual settings.
- Enter the prompt and image count. Add LoRAs if needed, setting Model and CLIP strengths independently.
- Choose Generate in the Prompter. Its softly glowing outline follows generation across all requested images and reaches 100% when the workflow finishes. Preview or download saved results from Gallery.
New nodes and starter workflows use Manual with enhancement none. Enable the optional Krea enhancer only after installing its dependency. Existing workflows retain their saved sampling settings and selected legacy preset. Models and LoRAs remain subject to their own licenses.
To wire nodes yourself, drag the native output dot to a matching input dot: Prompter Prompt → Engine Prompt (request), LoRA Loader LoRAs → Engine LoRAs, and Engine Images → Gallery Images. Engine Report → Gallery Report adds generation metadata. The captions identify the existing ports; dragging still uses ComfyUI's dots. Use ComfyUI's native node controls to reveal optional advanced inputs; see advanced connections.
Behavior that matters
- LoRAs are applied in their displayed order, before the Engine encodes both prompts. Disabled rows and rows with both strengths at zero are skipped; an active nonzero LoRA with zero effective patch matches raises an error. “Loaded” reports registered patches, not a completed image-quality test. See advanced conditioning when encoding text upstream.
- Image count is handled by the backend. Images use
seed,seed + 1, and so on, modulo 2^64. The report stores the exact seeds as decimal strings. - Prompter progress follows native sampler updates, completed image decoding and workflow completion. Loading has no invented percentage; errors and cancellation never display successful completion. Optional connected text is still supported.
- Gallery IDs are serialized in workflows. The new package uses separate storage under
output/DMAI-NODES; existing DMAI Studio history is not deleted or automatically imported. - ZIP exports are limited to 100 images / 2 GiB. Full-history selection uses a snapshot so later images are not silently added.
- Invalid saved JSON remains visible for correction. It is not replaced by defaults.
- Preset upload is available in DMAI Enhanced. The preset export button has been removed; keep the original JSON file for reuse. Historical built-in presets remain available to saved workflows, with imported settings shown for new selections.
- This is a new package with new node IDs. Existing Studio, Carousel and Krea Edit workflows remain in their original package.
Compatibility
Built against ComfyUI v0.37.0, with the frontend supplied by that tag. Renderer and inference coverage are recorded in VALIDATION.md. Native connection ports remain available. External socket captions target Canvas; custom positioning and captions may differ in Nodes 2.0.
No cloud service or paid generation is enabled by installation. The HTTP routes use ComfyUI's existing access boundary; put authentication and TLS in front of a remote deployment.
Development
Run these commands from a Git checkout of DMAI NODES, with Node.js 22 or newer for JavaScript tests. The install ZIP does not include the test suites.
python -m unittest discover -s tests -v
node --test tests-js/*.test.mjs
python tools/build_release.py
The optional real-Comfy CPU LoRA comparison is described in docs/VALIDATION.md. Builds include only allowlisted source, docs, workflows and fonts.
MIT © 2026 DMAI / ceroomedia. Third-party notices.