CivArchive
    Krea 2 Character + Style: Two-LoRA Stack Workflow - v1.0
    NSFW
    Preview 148114874
    Preview 148114875
    Preview 148114876

    One Krea 2 graph with two LoRA slots: a character in the first, a style in the second. It is the stack I use for my own posts, with the strengths I measured. Core ComfyUI nodes only, no custom nodes.

    Both LoRAs are separate downloads. This page holds the graph only. Pick one character and one style from the lists below and put both files in models/loras.

    REQUIRED MODELS

    • Diffusion model: RedCraft 3.0 Krea 2, the INT8 convrot build, redcraft23_30Krea2_int8_convrot.safetensors, in models/diffusion_models. I tested these LoRAs on RedCraft only.

    • Text encoder: the Qwen3-VL 4B text encoder from the Krea 2 downloads (see the Krea 2 guide on comfyui-wiki.com); I run a community int8 ConvRot build of it (qwen3vl_4b_int8_convrot), and the bf16 or fp8 file works the same. It goes in models/text_encoders, loaded with CLIPLoader type krea2.

    • VAE: the Qwen-Image VAE, qwen_image_vae.safetensors, in models/vae, from Comfy-Org on Hugging Face.

    • Slot 1, one character LoRA: Shawy, version v3 (trigger shawyvoss), Zara Nyx (zaranyx) or Aria Solari (ariasolari).

    • Slot 2, one style LoRA: Gritty Ink (grittyink) or Shard Brush (facetposter).

    HOW IT IS WIRED

    • Diffusion model, then slot 1 (character), then slot 2 (style), then the KSampler. Both slots are LoraLoaderModelOnly, so there is no CLIP strength to set.

    • It opens with Zara at 0.8 and Gritty Ink at 0.8, the pair with the most tests behind it, and a prompt written for that pair.

    • To run one LoRA alone, set the other slot's strength to 0.

    STRENGTHS

    Character first, style second.

    • Tested: Zara 0.8 + Gritty Ink 0.8. Same seed at Gritty Ink 0.6 and 0.8: she held at both, and 0.8 gave more dry-brush. After that, five frames for a post and one contest entry, all on this pair.

    • Tested: Shawy 0.6 + Gritty Ink 0.8. Kept on the first render on v3 and on v2. On v3 the eyes needed a fix in post (see LIMITS). Use 0.6 when she faces the camera. In an ink style on v2, a rear view at 0.6 grew a second bow and held at 0.8.

    • Tested once, in the showcase, each kept on the first render: Zara 0.8 + Shard Brush 1.0, Shawy 0.8 + Shard Brush 1.0, Aria 0.8 + Gritty Ink 0.8, Aria 0.8 + Shard Brush 1.0. Each number is that LoRA's own best strength, and none needed a change.

    • Ranges on their own: characters 0.6 to 1.0, default 0.8. Gritty Ink 0.6 to 1.0, default 0.8. Shard Brush 0.8 to 1.0, and 1.0 for people. At 1.2 Shard Brush overcooks into specks.

    PROMPT ORDER

    1. Optional fix lines at the top, one per trait that drifts, like Her nails: ....

    2. The house sentence, then the style trigger and its line. For Gritty Ink: A painterly anime illustration in a soft cel-shaded semi-realistic register. grittyink A gritty black ink illustration in heavy brush strokes and dry-brush texture on white paper, drawn rather than photographed. For Shard Brush, facetposter goes in the same place. Then the place.

    3. The character, placement first: Standing in the doorway is zaranyx, ..., then the outfit and the face.

    4. The pose, where each hand is, and the light.

    5. Shard Brush only: end with one of its eight palettes, such as crimson and wine red with gold and saffron light.

    On Zara with Gritty Ink, the same seed without the house sentence let more colour in, so keep it.

    PER LORA

    • Zara: zaranyx, an adult woman of 34, with long wavy jet-black hair shot through with crimson streaks and deep blood-red eyes, and, in the outfit, a wire-wrapped raw amber pendant at her throat. No phrase, no pendant.

    • Shawy: shawyvoss, and always with a small pink bow pinned at one side of her hair. Keep a glove phrase and an adult build line. On v3, also describe her ears, eyes and skin, or raise her to 1.0: at 0.8 a short prompt can lose her.

    • Aria: ariasolari. Use 1.0 when the prompt is only the trigger and a scene.

    • Gritty Ink: name one accent colour. I give it to the character's own colour: Zara's streaks and eyes, Shawy's bow, Aria's hair. Write the rest of the scene in ink words: hatching, solid black, bare white paper. Colour nouns in the scene (red brick, warm gas flames, a bronze statue) kept my Zara test in full colour.

    • Shard Brush: end with a palette and give the clothes a colour. White and khaki clothes stay photo-like; black held in my showcase.

    SETTINGS

    • Steps 8. It is a distilled model, more steps do not help. CFG 1.0, sampler er_sde, scheduler simple, denoise 1.0, 1 megapixel sizes (896x1152, 1152x896, 1024x1024).

    • Decode: VAEDecodeTiled, tile 512, overlap 64. Negative: a ConditioningZeroOut, because a negative prompt does nothing at CFG 1.0.

    LIMITS

    • Shawy v3 can draw a white iris and claw tips on the gloves in ink and poster styles. Both paint out in post. My ink frame needed the eye fix and still has a few small claw tips; the rear-view Shard Brush frame needed nothing.

    • Ink words do not stop every leak. A few native colours still came through in my Zara frames: green leaves and yellow jars.

    • Tested with one character per frame. Other people's hands in a Shawy frame can come out wearing her glove.

    • The stack was tested on RedCraft 3.0 (Krea 2) on ROCm (a 7900 XTX), with this node chain. Core nodes only, so CUDA builds should run it as is.

    If this worked for you, a thumbs-up here helps the next person find it.

    Description

    One Krea 2 graph with two LoRA slots: a character in the first, a style in the second. It is the stack I use for my own posts, with the strengths I measured. Core ComfyUI nodes only, no custom nodes.

    Both LoRAs are separate downloads. This page holds the graph only. Pick one character and one style from the lists below and put both files in models/loras.

    REQUIRED MODELS

    • Diffusion model: RedCraft 3.0 Krea 2, the INT8 convrot build, redcraft23_30Krea2_int8_convrot.safetensors, in models/diffusion_models. I tested these LoRAs on RedCraft only.

    • Text encoder: the Qwen3-VL 4B text encoder from the Krea 2 downloads (see the Krea 2 guide on comfyui-wiki.com); I run a community int8 ConvRot build of it (qwen3vl_4b_int8_convrot), and the bf16 or fp8 file works the same. It goes in models/text_encoders, loaded with CLIPLoader type krea2.

    • VAE: the Qwen-Image VAE, qwen_image_vae.safetensors, in models/vae, from Comfy-Org on Hugging Face.

    • Slot 1, one character LoRA: Shawy, version v3 (trigger shawyvoss), Zara Nyx (zaranyx) or Aria Solari (ariasolari).

    • Slot 2, one style LoRA: Gritty Ink (grittyink) or Shard Brush (facetposter).

    HOW IT IS WIRED

    • Diffusion model, then slot 1 (character), then slot 2 (style), then the KSampler. Both slots are LoraLoaderModelOnly, so there is no CLIP strength to set.

    • It opens with Zara at 0.8 and Gritty Ink at 0.8, the pair with the most tests behind it, and a prompt written for that pair.

    • To run one LoRA alone, set the other slot's strength to 0.

    STRENGTHS

    Character first, style second.

    • Tested: Zara 0.8 + Gritty Ink 0.8. Same seed at Gritty Ink 0.6 and 0.8: she held at both, and 0.8 gave more dry-brush. After that, five frames for a post and one contest entry, all on this pair.

    • Tested: Shawy 0.6 + Gritty Ink 0.8. Kept on the first render on v3 and on v2. On v3 the eyes needed a fix in post (see LIMITS). Use 0.6 when she faces the camera. In an ink style on v2, a rear view at 0.6 grew a second bow and held at 0.8.

    • Tested once, in the showcase, each kept on the first render: Zara 0.8 + Shard Brush 1.0, Shawy 0.8 + Shard Brush 1.0, Aria 0.8 + Gritty Ink 0.8, Aria 0.8 + Shard Brush 1.0. Each number is that LoRA's own best strength, and none needed a change.

    • Ranges on their own: characters 0.6 to 1.0, default 0.8. Gritty Ink 0.6 to 1.0, default 0.8. Shard Brush 0.8 to 1.0, and 1.0 for people. At 1.2 Shard Brush overcooks into specks.

    PROMPT ORDER

    1. Optional fix lines at the top, one per trait that drifts, like Her nails: ....

    2. The house sentence, then the style trigger and its line. For Gritty Ink: A painterly anime illustration in a soft cel-shaded semi-realistic register. grittyink A gritty black ink illustration in heavy brush strokes and dry-brush texture on white paper, drawn rather than photographed. For Shard Brush, facetposter goes in the same place. Then the place.

    3. The character, placement first: Standing in the doorway is zaranyx, ..., then the outfit and the face.

    4. The pose, where each hand is, and the light.

    5. Shard Brush only: end with one of its eight palettes, such as crimson and wine red with gold and saffron light.

    On Zara with Gritty Ink, the same seed without the house sentence let more colour in, so keep it.

    PER LORA

    • Zara: zaranyx, an adult woman of 34, with long wavy jet-black hair shot through with crimson streaks and deep blood-red eyes, and, in the outfit, a wire-wrapped raw amber pendant at her throat. No phrase, no pendant.

    • Shawy: shawyvoss, and always with a small pink bow pinned at one side of her hair. Keep a glove phrase and an adult build line. On v3, also describe her ears, eyes and skin, or raise her to 1.0: at 0.8 a short prompt can lose her.

    • Aria: ariasolari. Use 1.0 when the prompt is only the trigger and a scene.

    • Gritty Ink: name one accent colour. I give it to the character's own colour: Zara's streaks and eyes, Shawy's bow, Aria's hair. Write the rest of the scene in ink words: hatching, solid black, bare white paper. Colour nouns in the scene (red brick, warm gas flames, a bronze statue) kept my Zara test in full colour.

    • Shard Brush: end with a palette and give the clothes a colour. White and khaki clothes stay photo-like; black held in my showcase.

    SETTINGS

    • Steps 8. It is a distilled model, more steps do not help. CFG 1.0, sampler er_sde, scheduler simple, denoise 1.0, 1 megapixel sizes (896x1152, 1152x896, 1024x1024).

    • Decode: VAEDecodeTiled, tile 512, overlap 64. Negative: a ConditioningZeroOut, because a negative prompt does nothing at CFG 1.0.

    LIMITS

    • Shawy v3 can draw a white iris and claw tips on the gloves in ink and poster styles. Both paint out in post. My ink frame needed the eye fix and still has a few small claw tips; the rear-view Shard Brush frame needed nothing.

    • Ink words do not stop every leak. A few native colours still came through in my Zara frames: green leaves and yellow jars.

    • Tested with one character per frame. Other people's hands in a Shawy frame can come out wearing her glove.

    • The stack was tested on RedCraft 3.0 (Krea 2) on ROCm (a 7900 XTX), with this node chain. Core nodes only, so CUDA builds should run it as is.

    Workflows
    Krea 2

    Details

    Downloads
    48
    Platform
    CivitAI
    Platform Status
    Available
    Created
    11/6/2026
    Updated
    11/11/2026
    Deleted
    -

    Files

    krea2-character-style.zip

    Mirrors

    CivitAI (1 mirrors)