Krea 2 Medium
The cheaper tier for testing prompts against the hype model
- referenceImages
- advancedFeatures.watermark.image
- image
Krea 2 Medium is the budget version of the image model everyone was talking about in 2026. Same architecture family, same interface, lower price tier - which makes it the right place to find out whether Krea 2's aesthetic actually fits your project before you spend Large-tier credits on it. It's the image-node equivalent of testing a video prompt on Kling Standard before committing to Pro.
The node runs the model on Runware's cloud, so the whole "Krea 2 is heavy to run locally" problem evaporates. No checkpoint, no Qwen3-VL text encoder, no VAE - the pack sends the request and hands you back a native IMAGE tensor. Keep in mind this is the hosted Krea 2, not the open Krea-2-Raw weights, and Krea has confirmed the open and hosted versions aren't identical. For iterative work that rarely matters; for shipping, know what you're renting.
How it works
Same thin glue as every node in ComfyUI-Runware: the widgets become a request for model AIR krea:krea@2-medium with an imageInference task, the runware-sdk sends it over REST, and the image comes back as an IMAGE output for Preview Image or Save Image.
The inputs mirror the Large tier exactly:
positivePrompt- natural language, long and specific; wrap in-image text in quotes.width/height- required, 1024 default.settings.creativity-raw,low,medium,high, defaultmedium.rawis the undistilled Raw behavior; lower values hug the prompt, higher values let the model improvise.seed- reproducibility.numberResults- 1–20 variations; keep it low on the budget tier.advanced_json- forsettings.moodboardswhen you want style steering.
There's also a watermark socket and the usual safety gate (off by default - the model's own default applies).
The output
One image socket, type IMAGE. Default output is JPG at quality 95; PNG for lossless. Nothing else to do.
Installing it
The pack install is a one-time thing for all these nodes:
cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt
Restart, set your key in ComfyUI Settings → Runware API key (or RUNWARE_API_KEY, or runware auth login), and you're generating. Dependencies are light - SDK, pillow, soundfile - with zero model downloads.
Where people get burned
Judging Krea 2 by Medium renders, for one. Medium is the iteration tier; if the prompt idea is solid, the Large tier is where the detail lands. The other trap is forgetting this is a different model from the open weights - if you're tuning a prompt here and porting it to a local Krea-2-Raw setup, expect drift. And numberResults is the quiet credit-burner: twenty variations of "hmm, close" is a real receipt.
Inputs (21)
| Name | Type | Default | Description |
|---|---|---|---|
| positivePrompt | STRING | Text prompt describing elements to include in the generated output. | |
| referenceImagesopt | IMAGE | — | |
| advancedFeatures.watermark.imageopt | IMAGE | — | |
| sizeopt | COMBO | 1K (1:1) | Output size. Only one of resolution or dimensions is sent. |
| seedopt | INT | 00–9223372036854776000 | Random seed for reproducible generation. When not provided, a random seed is generated in the unsigned 32-bit range. |
| numberResultsopt | INT | 11–20 | Number of results to generate. Each result uses a different seed, producing variations of the same parameters. |
| advancedFeaturesopt | BOOLEAN | false | Enable to set advancedFeatures. Off uses the model's default. |
| advancedFeatures.watermark.bgColoropt | STRING | Background color in hex format. | |
| advancedFeatures.watermark.displayPositionopt | COMBO | (default) | Watermark position. |
| advancedFeatures.watermark.fontColoropt | STRING | Text color in hex format. | |
| advancedFeatures.watermark.opacityopt | BOOLEAN | false | Enable to set advancedFeatures.watermark.opacity. Off uses the model's default. |
| advancedFeatures.watermark.opacity_valueopt | FLOAT | 0.100.1–1 | Watermark opacity. |
| advancedFeatures.watermark.textopt | STRING | Watermark text. | |
| settings.creativityopt | COMBO | medium | Creativity level. Controls how closely the model follows the prompt versus generating its own interpretation. |
| safetyopt | BOOLEAN | false | Enable to set safety. Off uses the model's default. |
| safety.checkContentopt | BOOLEAN | false | Enable or disable content safety checking. |
| ttlopt | BOOLEAN | false | Enable to set ttl. Off uses the model's default. |
| ttl_valueopt | INT | 60 | Time-to-live (TTL) in seconds for generated content. Only applies when `outputType` is `URL`. |
| outputFormatopt | COMBO | JPG | File format for the generated image. |
| outputQualityopt | INT | 9520–99 | Compression quality of the output. Higher values preserve quality but increase file size. |
| advanced_jsonopt | STRING | Optional JSON merged into the request. For: settings.moodboards |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |