Six adjustable LoRA sliders for playful, imperfect photography. Each version controls one effect:
• Yellow-Green Color Cast — a warm yellow-green tint.
• Direct Flash — a direct, on-camera flash look.
• Defocus — missed-focus softness and diffuse edges.
• Motion Trails — motion blur and trailing softness.
• Fisheye — curved wide-angle geometry.
• Honeycomb Prism — repeated views inside refractive polygonal cells.
For Krea 2. No trigger words required. Load like a normal LoRA and adjust model strength; 0 is off. In ComfyUI, use Load LoRA (model only), or a regular LoRA loader with CLIP strength 0. Use one effect or combine several.
Start with Color Cast 2, Direct Flash 2, Defocus 5, or Motion Trails 3. Available slider ranges: 0–3 / 0–3 / 0–10 / 0–6. Strong portrait examples use Defocus 10 or Motion Trails 4.5–6. These are scene-dependent: higher weights can also change facial details, clothing, pose and composition.
Includes a ComfyUI workflow with draggable sliders and 4-, 6-, or 9-photo collages. Each photo has its own prompt, seed and four controls: use no effects, one effect, or stack several. Requires Goohaitools-ComfyUI and ComfyUI-Easy-Use. All three collage sizes were tested.
Tested with Krea 2 Turbo BF16, 8 steps, CFG 1, Euler ancestral / beta. Each version has a young adult portrait strength comparison using the same prompt and seed. No effect words or post-processing blur were added to the comparisons.
Standalone collage workflow: Download the 4/6/9-photo workflow JSON and ZIP. See the nine-photo result and workflow-controls screenshot.
New lens effects: Fisheye — curved wide-angle geometry, recommended 3–3.5, start 3.5. Honeycomb Prism — repeated photographic views inside glass-edged polygonal cells, recommended 3–4, start 3.5. Both use ordinary single-pass 8-step sampling with the selected LoRA active throughout. Tested with official Krea 2 Turbo BF16, 768 × 1344, CFG 1, Euler ancestral / beta. Adjust model strength directly; no trigger words or special sampling schedule are needed. Faces, clothing, poses, surface details and framing can change. The single-pass workflow ZIP and metadata PNG examples are included.
Description
Honeycomb Prism repeats the subject and surroundings inside glass-edged polygonal cells, with refracted borders and subtle rainbow highlights.
For Krea 2. Start at strength 3.5; 0 is off. No trigger words are required. Use an ordinary scene description and a model-only LoRA loader (or set CLIP strength to 0).
Tested settings: official Krea 2 Turbo BF16, 768 × 1344, 8 total steps, CFG 1, Euler ancestral / beta. Text encoder: qwen3vl_4b_fp8_scaled.safetensors. VAE: qwen/qwen_image_vae.safetensors. Examples use no face restoration or detailer.
Recommended strength: 3–4; ordinary experimentation range 2.5–4. Around 2.5 the repetition becomes apparent; 3–4 forms fuller optical cells. Apply the LoRA across all 8 steps. Set Fisheye to 0 when using Honeycomb Prism alone.
4.5–5 is an abstract overdrive range: faces and bodies can fragment or disappear. At 3.5 the effect appears across all six new portraits, but it can change faces, apparent age, clothing, pose and framing. Some cells contain only background or partial subjects.
FAQ
Comments (2)
This is the most genius and realistic LoRA I've ever seen.
thanks! Follow me please！hahaha