Hybrid ANIMA KREA2 ComfyUI Workflow

Hybrid ANIMA_KREA2 ComfyUI Workflow This is a hybrid ComfyUI workflow that uses ANIMA for character and composition, then uses KREA2 as an img2img refinement pass for details, text, and sharpness. The basic idea is simple: ANIMA creates the base image KREA2 refines the image with better details and text rendering FaceDetailer can optionally clean up the face at the end This workflow is useful if you like ANIMA’s character knowledge and composition control, but also want to take advantage of KREA2’s stronger detail rendering and text-handling ability. About the Sample Image The sample image was originally intended to feature an elf of around 1,090 years old. However, since humans and AIs have very limited ways to determine an elf’s age, the actual sample subject is a human who is approximately 1,000 years younger than the elf originally planned. Workflow Structure Stage 1: ANIMA Base Image ANIMA is used to create the first image. This stage is mainly responsible for: character identity pose outfit structure overall composition general scene layout Main files used: Anima_liquidmix_base_v1.12.safetensors qwen_3_06b_base.safetensors qwenimageVAE_liquid1127.safetensors The workflow also includes an optional ANIMA LoRA loader. Stage 2: KREA2 img2img Refinement The ANIMA image is re-encoded with VAEEncode , then sent into KREA2 as an img2img input. This stage is mainly responsible for improving: fine details printed text small accessories patterns edge clarity overall sharpness Main files used: liquidmix_krea2_turbo_fp8_v0.22.safetensors qwen3vl_4b_bf16.safetensors qwenimageVAE_liquid1127.safetensors The default KREA2 denoise value is: 0.35 Recommended denoise values: 0.25 — preserves the ANIMA image more strongly 0.35 — balanced default 0.45 — stronger KREA2 changes and detail correction If the image changes too much, lower the denoise value. If the KREA2 pass is too weak, raise it slightly. Optional Stage 3: FaceDetailer The final optional stage uses FaceDetailer to clean up the face after the KREA2 refinement pass. This stage uses: KREA2 model and CLIP YOLO face detection FaceDetailer from Impact Pack The FaceDetailer result is saved separately from the normal KREA2 refined image, so you can compare both versions. Preview and Comparison This workflow includes nodes for checking the intermediate results. You can preview or compare: the ANIMA base image the KREA2 refined image the final image after FaceDetailer There is also an Image Comparer node for comparing the ANIMA output and the KREA2 output. Please note: The workflow notes may mention a 3-panel comparison image, but this JSON does not include the full node group needed to save a side-by-side 3-panel image. To save a 3-panel comparison image, add nodes such as EmptyImage , ImageCompositeMasked , and SaveImage , then place the ANIMA / KREA2 / FaceDetailer images side by side. Required Model Files Place the files in the following folders: ComfyUI/ └── models/ ├── diffusion_models/ │ ├── Anima_liquidmix_base_v1.12.safetensors │ └── liquidmix_krea2_turbo_fp8_v0.22.safetensors ├── text_encoders/ │ ├── qwen_3_06b_base.safetensors │ └── qwen3vl_4b_bf16.safetensors ├── vae/ │ └── qwenimageVAE_liquid1127.safetensors └── loras/ ├── plush_AN_basev1.safetensors └── plush_Krea2_v1.safetensors The LoRA files are optional. Set the LoRA strength to 0 or select [none] if you do not want to use them. Required Custom Nodes This workflow may require the following custom nodes: comfyui-impact-pack comfyui-impact-subpack rgthree-comfy comfyui-custom-scripts comfyui-krea2-conditioning When to Use This Workflow Use this workflow if you want: ANIMA-style character and pose control KREA2-style detail refinement better text and small-object rendering img2img refinement while keeping the original composition a way to compare ANIMA and KREA2 behavior a test workflow for LoRAs across both models This is mainly an experimental bridge between ANIMA and KREA2. ANIMA handles the first draft and composition. KREA2 handles detail refinement and text improvement. FaceDetailer is used only as an optional final cleanup step. Because this workflow loads multiple large models, VRAM usage may be high. If you run into memory issues, try disabling optional LoRAs, lowering the resolution, or turning off the FaceDetailer stage.