Krea 2 · self-hosted
Krea 2 images model, self-hosted on Wavemaker's GPU fleet. Pick it in the model picker or name it in your brief. This page gathers its model configurations and example graphs.
Discover · Wavemaker catalog
When to use it
- Krea 2 images model, self-hosted on Wavemaker's GPU fleet. Pick it in the model picker or name it in your brief. This page gathers its model configurations and example graphs.
What you provide
- A written scene brief and desired image dimensions.
- A precomputed control map compatible with an operator-installed custom adapter.
- The exact adapter identity and requested image brief.
- An owned image whose depth structure should guide the result.
What you get
- An editable configuration and bound Comfy graph; rendered media is produced only after a separate run.
- An explicit krea model configuration and example graph.
- One still at the scene plate's aspect, at most 2 MP.
- One edited still at the canvas size, at most 2 MP.
Creative guidance
- Krea 2 · native Turbo still: Generate a still with native Krea 2 Turbo at 768×1344; the legacy seed race is replaced by reviewing and rerunning seeds.
- Krea 2 · custom control map: Provide a precomputed control map and select the operator-installed Krea 2 Control-LoRA trained for it; the node alone does not establish an adapter.
- Krea 2 · depth-guided image: Keep an input image's depth structure: Depth Anything 3 depth conditions Krea 2 Turbo through the depth Control-LoRA.
- Krea 2 · source-image edit: Restage an owned source image with the native Krea 2 identity edit (identity LoRA, in-context source tokens and image-grounded instruction).
- krea2 edit multi: krea model configuration for image editing.
- krea2 edit native: krea model configuration for image editing.
- krea2 edit pair native: krea model configuration for image editing.
- Krea2 person into a scene · native: Restage one person into the same scene plate pose after pose: the room, furniture and light stay fixed and the person is relit by the scene.
- Krea2 General Edit · native: Instruction edits of one image in the Krea2 look, keeping what the instruction does not mention.
- Krea 2 · H100 service: Select the explicitly configured H100 Krea 2 service rather than the separate native Turbo graph.
- krea2 inpaint: krea model configuration for image editing.
- krea2 mrflow: krea model configuration for image editing.
- krea2 native raw: krea model configuration for text-to-image.
- Krea 2 Turbo · native Comfy: Use the native eight-step Turbo graph with its exact Qwen encoder and VAE; service, Raw, edit and control profiles remain separate.
- krea2 outpaint: krea model configuration for image editing.
- Krea 2 · pose-control image: Provide a precomputed pose map to Krea 2 Turbo through the pose Control-LoRA; an ordinary photo is not the required control map.
- Krea 2 · Raw with Turbo LoRA: Choose the Raw transformer with the Turbo-distill LoRA at 0.6 and twelve steps; it differs from Turbo and from the two-stage graph.
- Krea 2 · Turbo: Choose the explicit Krea 2 Turbo checkpoint: eight steps at CFG 1, one seed per run.
- Krea 2 · Raw then Turbo: Run eight Raw steps at CFG 3.5, then four Turbo steps over fixed sigmas on the Raw latent; it is not a switchable refinement.
- krea2 wan2x decode: krea model configuration for text-to-image.
- Community shapes: Licensed looks ported from community harvests — our graphs, not copied workflows.
- Gear · In-house: Self-hosted image and video path — LTX stills and motion on our GPUs.
- Model · LTX hybrid: Self-hosted LTX stills and video.
- Krea 2 depth ControlNet: Depth-conditioned Krea 2 still. Attach a control image so pose and layout follow the reference.
- Krea2 General Edit: A Comfy graph that edits one image from a plain instruction (recolor, relight, season, sign text, add or remove objects) with Krea2 and the General Edit V2 LoRA.
- Krea 2 identity edit: Identity-preserving edit. Upload a still and describe the change — the person stays the same person.
- Krea2 identity restage: A Comfy graph that restages the person in one photo into a new scene, pose, outfit and light with Krea2 Turbo and the identity edit V2 LoRA.
- Krea2 two-person restage: A Comfy graph that places two people from two photos into one new shot with Krea2 Turbo and the identity edit v1.2 LoRA.
- Krea2 person into a scene: A Comfy graph that places the person from one photo into a fixed scene plate, keeping the room and its light, for consistent keyframes across poses.
- Krea2 identity restage (v1): The v1 Comfy graph that model krea2-edit runs in production: Krea2 Turbo and the identity edit v1 LoRA with crop geometry restage the person in one photo into a new scene.
- Hosted Krea-2 Turbo text-to-image (engine): Hosted Krea-2 Turbo text-to-image (engine). Authored Denoise Engine node; standard rail.
- Krea 2 scene with two people or a product: A Comfy graph that puts two people, or a person and a product, into one scene plate in a single Krea 2 job: two identity-edit passes that keep the room, its light and each face from its reference.
- Krea 2 region repair (inpaint): A Comfy graph that repairs or changes one masked region of a Krea 2 still with the krea2-anypaint adapter on a 1024 crop and stitches it back, so the rest of the picture keeps its pixels and its Krea look.
- Krea 2 outpaint to a new frame: A Comfy graph that fits a still into a larger canvas at an anchor and generates the rest in the Krea 2 look with the krea2-anypaint adapter, keeping the original pixels beyond a narrow feather.
- LTX Best Seed (hybrid): LTX best-seed hybrid pipeline — races multiple seeds on the self-hosted LTX stack and keeps the strongest result. Tuned for quality-per-credit.
Setup and requirements
- Name the option you want in your brief; the agent reads its method or configuration (Source entry points) and applies it.
Source licenses and model terms continue to apply. Check runtime availability before generation.
Download source · Learn how to reuse resources