Aller au contenu
WorkflowPrésentéExterne
Télécharger

Krea2 Turbo Workflow + Upscale + Reference + Editing + Uncensored — v2.1 — v4.0 Revamped

This comparison uses the original V3 JSON supplied by the author, not an

VersionRelease 3309925 — professionalKrea2_v40Revamped.jsonEstimé VRAMNot yet characterizedUnratedClasse

Profil d'exécution

Accès
Flux de travail externe
Statut
Présenté
Version
professionalKrea2_v40Revamped.json
Source
Civitai
Mis à jour
4 nov. 2026

Description de la source

Donut Workflow changelog

V4 Beta — compared with the original V3

This comparison uses the original V3 JSON supplied by the author, not an

intermediate development file that was also named V3. The source fingerprints

and structural comparison are recorded in comparison.json.

Redesigned interface

  • Replaced the spread-out loader/settings layout and rgthree labels with

numbered Donut control cards and collapsible Advanced sections.

  • Moved source loaders and generation internals into inspectable subgraphs.

  • Added direct block-weight slider controls, automatic card sizing and shared

controls across Graph and App Mode. LoRA stacking and ordering existed in V3.

  • Added a consolidated latest-result panel, stage progress and final expanded

prompt display while retaining the individual stage outputs.

Images and editing

  • Replaced the separate Load Image, editing toggle, edit-LoRA loader and

resolution selector with Edit Studio .

  • Added two image slots: A for the base/scene and optional B for subject/identity,

including B-based face-reference routing.

  • Added paste/drop/upload, visual crop controls and integrated preset/custom,

reference-aspect and crop-aligned sizing.

  • Added Reference guidance as a separate native image-conditioning path for

generation, without an edit LoRA. It pauses while Editing is enabled.

V3 already had a single-image edit path and a face detailer. These changes are

editing/control additions, not a claim of universally smarter face detailing.

Prompts and guidance

  • Added a persistent Wildcard library , picker and expanded-text preview;

explicit wildcard tokens replace the old automatic prompt-addition controls.

  • Added shared seed-variance controls for general and face positives, including

reapplication to freshly encoded edit conditioning.

  • Integrated NAG controls into the sampler, upscale and face-detail paths,

including edit-mode negative conditioning. V3 already used a standalone NAG

node; NAG itself is not new to the workflow.

Installation and models

  • Added Download missing with a reviewed model catalog, size/hash checks,

reuse of matching local files and cancellation.

  • Added a collapsed Required node packs subgraph so Manager can discover

internally called Krea2 Edit, NAG and seed-variance dependencies.

  • Fixed DonutFaceDetailer registration depending on Impact Pack loading first.

Dependency loading failures are distinct from image-quality improvements.

  • Removed WAS and rgthree label requirements from this workflow. Seven companion

packs remain discoverable through the normal missing-node installer.

Saving

  • Replaced WAS Image Save with Donut Image Save , without adding dependencies.

  • Preserved the active Final folder, seed-based filenames, numbering, WebP

quality, overwrite setting and secondary core save output.

  • Retained MIT attribution for the adapted WAS filename numbering code.

  • Removed unused WAS-only history, color-profile and high-bit-depth/EXR controls.

Defaults and compatibility

  • Existing V3 concepts remain: model merging/Fusion Control, LoRA stacking,

Turbo sampling, face detailing, two upscale stages and seed planning.

  • The beta is a new preset as well as a new layout; saved settings differ from

the original V3. Carry personal settings over explicitly rather than assuming

the same prompt/seed will reproduce a V3 image.

  • The distributed configuration has editing off, first upscale on and second

upscale off. It includes optional seed variance and integrated NAG controls.

  • Requires the updated DonutNodes code shipped with this workflow. A node-pack

release number is not the workflow's V4 version number.

Validation

  • Fresh default missing-node installation, restart and full generation passed

with the development DonutNodes code; no WAS selection or repair override.

  • Verified base generation, first upscale, two face refinements and a 1728 × 1344

WebP save. Output matched the preceding WAS save for the same tested preset.

  • Six saver tests and 34 registration/dependency-isolation tests passed.

  • Editing, the disabled second upscale, model downloads and Windows/macOS were

not exercised by that final fresh-install run. These remain beta test areas.

What you need

Model files
4.9 GiB known subtotal

Remaining hardware or download-size estimates are not available yet.

Exigences

Modèles

9
  • 4x_NickelbackFS_72000_G.pth

    Non résolu

    Upscaler · Unknown

  • face_yolov8m.pt

    Non résolu

    Object Detector · Unknown

  • Non résolu

    Diffusion model · Unknown

  • krea2_identity_edit_v1_2.safetensors

    Non résolu

    Checkpoint · Unknown

  • krea2_turbo_bf16.safetensors

    Non résolu

    Diffusion model · Unknown

  • qwen_image_vae.safetensors

    Possible

    VAE · Unknown

  • Text encoder · 4.88 GB · SAFETENSORS · U8 + FP32 + BF16 + FP8 (tensor header) · Unknown

  • sam_vit_b_01ec64.pth

    Dépôt identifié

    Segmentation · Unknown

  • Unresolved model selection: DonutModelDownloads

    Non résolu

    Unknown

Packs de nœuds

9
  • ComfyUI-bleh

    Trouvé

    Pack de nœud · Registry

  • comfyui-impact-pack

    Trouvé

    Pack de nœud · Registry

  • Pack de nœud · GitHub · ltdrdata/ComfyUI-Impact-Subpack

  • derfuu_comfyui_moddednodes

    Trouvé

    Pack de nœud · Registry

  • DonutLatestPreview

    Non résolu

    Pack de nœud · Unknown

  • DonutModelDownloads

    Non résolu

    Pack de nœud · Unknown

  • donutnodes

    Trouvé

    Pack de nœud · GitHub · DonutsDelivery/ComfyUI-DonutNodes

  • DonutReferenceStudio

    Trouvé

    Pack de nœud · Unknown

  • DonutWorkflowPanel

    Non résolu

    Pack de nœud · Unknown

Commentaire généré par l’IA

Explication générée par l’IA à partir des détails de la source et de la configuration. Les suggestions sont clairement signalées.

Objectif

  • This Krea 2 workflow generates images from text prompts and supports image reference or editing, face refinement, and upscaling; it takes prompt text plus optional base and subject images and produces stage previews and saved WebP images.

Flux

  • The panels show stage progress and the latest result, with individual stage outputs and an expanded final prompt.

Entrées et sorties

  • Edit Studio has image slots A for the base scene and optional B for a subject or identity, with B also available for face-reference routing. You can paste, drop, or upload images, then crop and size them.

  • Donut Image Save supports the active Final folder, seed-based filenames, numbering, WebP quality, an overwrite setting, and a secondary core save output.

Installation

  • The workflow requires the updated DonutNodes code shipped with it.

Dépendances

  • The workflow lists these node packs: ComfyUI-bleh, comfyui-impact-pack, comfyui-impact-subpack, derfuu_comfyui_moddednodes, and donutnodes.

  • Listed model files include krea2_turbo_bf16.safetensors, finepornV4INT8NVFP4BF16_v4.safetensors, krea2/krea2_identity_edit_v1_2.safetensors, and qwen3vl_4b_fp8_scaled.safetensors.

  • Other listed model files are qwen-image/qwen_image_vae.safetensors, 4x_NickelbackFS_72000_G.pth, bbox/face_yolov8m.pt, and sam_vit_b_01ec64.pth.

Configuration

  • The distributed settings start with Editing off, the first upscale on, and the second upscale off; seed variance is optional, and NAG controls are integrated.

Précautions

  • This V4 release uses different saved settings from the original V3; carry your settings over explicitly instead of expecting the same prompt and seed to reproduce a V3 image.

    Suggestion

  • The reported fresh-install run verified image generation, first upscale, two face refinements, and a 1728 × 1344 WebP save; it did not exercise editing, the second upscale, model downloads, or Windows and macOS.

Faut-il modifier cela ?

Connectez-vous pour envoyer une demande de modification.

Sources

1 source
CivitaiPrincipalVersion 3309925
civitai.com/models/2169381

Vérifié 4 nov.

Workflow instructions

Author notes from professionalKrea2_v40Revamped.json · selected release

face_yolov8m.pt - URL: https://huggingface.co/Bingsu/adetailer/resolve/main/face_yolov8m.pt - Size: 52 MB - SHA256: 717923c19b3f4bbf5250b728f1fa6b2cb72a33aed1d236ea9caf0e21ad943e5f - Folder: ComfyUI/models/ultralytics/bbox/ nickelback → 4x_Nickelback_70000G.pth (ESRGAN upscaler) - URL: https://huggingface.co/licyk/sd-upscaler-models/resolve/3a2094a57f456c41b79e5a4d196f2222c9940b89/ESRGAN/4x_Nickelback_70000G.pth?download=true - Size: 67 MB - Folder: ComfyUI/models/upscale_models/ sam_vit_b_01ec64.pth - URL: https://dl.fbaipublicfiles.com/segment_anything/sam_vit_b_01ec64.pth - Size: 375 MB - SHA256: ec2df62732614e57411cdcf32a23ffdf28910380d03139ee0f4fcbe91eb8c912 - Folder: ComfyUI/models/sams/ zImageClearVae_halfNatural.safetensors - URL: https://huggingface.co/easygoing0114/Z-Image_clear_vae/resolve/main/Z-Image_half_natural_vae.safetensors - Size: 335 MB - Folder: ComfyUI/models/vae/ - Note: HF filename is Z-Image_half_natural_vae.safetensors (underscores); the receiver would need to rename to zImageClearVae_halfNatural.safetensors to match the workflow reference, or we ship an alias. - Also has variants in the same repo: natural, clear, vivid.

Download files

Artefacts du modèle

2 artefacts

professionalKrea2_v40Revamped.json

json · 277 KB · SHA-256 497dce0dbc9e…4b0b · Civitai

Download file

professionalKrea2_v40Revamped.json

json · 277 KB · SHA-256 497dce0dbc9e…4b0b

Download file

Extraits de sources

2 extraits

Source context: 33033 downloads · Type Workflows · Base model Krea 2

Source context: 31437 downloads · Type Workflows · Base model Krea 2