Krea 2 Muse text-to-image ComfyUI workflow#
This Krea 2 Muse text-to-image ComfyUI workflow is a RunComfy-ready, one-click image generation pipeline for portraits, square visuals, and wide landscapes. Built around a centralized sampler, it keeps size, seed, guidance, and prompt controls in one place so you can explore variations quickly.
Designed for concept art, editorial imagery, product concepts, and campaign ideation, the workflow pairs the speed of Krea 2 Turbo with a clean prompt-entry path and a labeled preview grid for fast A/B testing. The default checkpoint targets a “Muse” aesthetic via a third‑party Krea 2 Turbo fine‑tune, while remaining fully compatible with official Krea 2 weights.
Key models in Comfyui Krea 2 Muse text-to-image ComfyUI workflow#
- Krea 2 Turbo. The core text-to-image model that powers this workflow, optimized for fast, high-quality sampling and broad aesthetic range. Open weights and variants are available on Hugging Face under the official organization krea/Krea-2-Turbo.
- Qwen‑series vision-language text encoder. Krea 2 packaging for ComfyUI ships with a Qwen VL encoder to parse prompts with strong text understanding, improving instruction following and visual detail. See the packaged text encoders in Comfy-Org/Krea-2 and background docs in the Diffusers QwenImage pipeline.
- Qwen Image VAE. The autoencoder used to compress and reconstruct images, balancing fidelity and speed for Krea 2 family pipelines. Reference implementation and files are published in Qwen/Qwen-Image.
How to use Comfyui Krea 2 Muse text-to-image ComfyUI workflow#
The graph follows a straightforward path from prompt assembly to a centralized sampler, then to preview and save. Prompts can be sourced from built‑in lists or typed freely, and all generation controls live on the sampler for fast iteration.
- Prompt library and prefix builder
- Use the curated prompt lists to jumpstart ideation across portraits, nature, wildlife, product scenes, cinematic moments, and stylized art. Each list is a collection of ready-to-run prompts you can adapt to your brief.
- The prefix node lets you add short style tokens or LoRA triggers that should consistently appear before your main prompt. This is ideal for enforcing a mood, medium, or project tag across runs.
- Connect your chosen prompt list to the concatenation node to merge the prefix and the main description, then route that text into the sampler’s positive prompt.
- Centralized sampling and controls
- The sampler is the heart of the Krea 2 Muse text-to-image ComfyUI workflow. It loads the Krea 2 Turbo checkpoint, the Qwen text encoder, and the Qwen Image VAE, then executes generation with your prompt.
- Set image size for portrait, square, or landscape. Adjust the guidance (CFG) and step count to balance speed, adherence, and detail. Krea 2 Turbo is designed to work well at relatively low guidance compared with classic diffusion models.
- Choose a sampler and scheduler appropriate for quick look‑dev or higher‑fidelity passes. You can also batch images to explore multiple seeds in one click.
- The sampler supports LoRA injection so you can introduce style packs or project‑specific fine‑tunes without restructuring the graph.
- Seed management for repeatability
- A dedicated seed node provides stable, reproducible results. Lock a seed to refine a single composition, or randomize to explore new arrangements and lighting before you commit.
- Preview and save
- A labeled comparison grid collects your outputs for quick side‑by‑side review. Use it to check framing, color, and fine details across seeds or parameter tweaks.
- The preview node displays results inline while the save node writes finals with a custom filename prefix for clean versioning and retrieval.
Key nodes in Comfyui Krea 2 Muse text-to-image ComfyUI workflow#
BinyuanUltimateSampler(#6) The central generator that loads the Krea 2 Turbo checkpoint, text encoder, and VAE, then handles size, steps, guidance, seed, sampler, and scheduler in one place. Use it to switch between portrait, square, and landscape outputs quickly. Keep guidance modest for Krea 2 Turbo, increase steps when chasing micro‑detail, and use batches to explore multiple seeds efficiently. If you apply LoRAs, adjust their strength conservatively to preserve base‑model coherence.CR Text(#8) A simple text field for prefix tokens that should always lead your prompt, such as a style tag, brand tag, or LoRA trigger. Keep prefixes short and separated by commas for clean conditioning.CR Text Concatenate(#13) Merges the prefix with your main prompt into a single, well‑formed description for the sampler. Use it to enforce consistent phrasing and to test how different prefixes steer the same base prompt.Seed (rgthree)(#7) A utility node for generating, locking, or shuffling the seed value that drives randomness. It is the quickest way to reproduce an image later or branch a set of controlled variations.binyuanMultiImageCompare(#3) Builds a labeled grid from your outputs for fast A/B review. Use it to annotate runs with a short tag, then evaluate composition, lighting, and texture fidelity across seeds or parameter changes.SaveImage(#20) Saves the rendered images to disk with a clear filename prefix. Name runs by project, orientation, or date to keep large explorations organized.
Optional extras#
- Orientation first, then style. Pick your target aspect ratio and composition early, then refine style with prefixes, LoRAs, and prompt edits. This keeps framing stable while you iterate on look.
- Keep guidance low with Krea 2 Turbo. Compared with classic diffusion models, this family responds well to modest CFG; nudge it upward only if the image drifts too far from the prompt.
- Use small batches for ideation. Generate a few seeds at once, shortlist in the comparison grid, then lock one seed for polishing.
- Licensing and safety. The official Krea 2 packaging includes the community license; review it before commercial use and apply reasonable safety measures as required. See the license in Comfy-Org/Krea-2.
This Krea 2 Muse text-to-image ComfyUI workflow is purpose-built for rapid visual exploration: one prompt in, consistent controls, clean previews out. Whether you are building concept boards, editorial treatments, or campaign frames, it gives you a fast, reliable foundation for turning ideas into images with Krea 2 Turbo.
Acknowledgements#
This workflow implements and builds upon the following works and resources. We gratefully acknowledge RunningHub for the krea2MuseByStable_v15TurboFp8 ComfyUI workflow, Stable Yogi for the Krea2 Muse model, Krea AI for the Krea 2 Open Source project, and Krea AI for the Krea 2 Turbo model weights for their contributions and maintenance. For authoritative details, please refer to the original documentation and repositories linked below.
Resources#
- RunningHub/krea2MuseByStable_v15TurboFp8
- Docs / Release Notes: krea2MuseByStable_v15TurboFp8
- Stable Yogi/Krea2 Muse
- Docs / Release Notes: Civitai model page
- Krea AI/Krea 2 Open Source
- GitHub: krea-ai/krea-2
- Hugging Face: krea/Krea-2-Raw
- Docs / Release Notes: Krea 2 Open Source page
- Krea AI/Krea 2 Turbo
- Hugging Face: krea/Krea-2-Turbo
- GitHub: krea-ai/krea-2
Note: Use of the referenced models, datasets, and code is subject to the respective licenses and terms provided by their authors and maintainers.