AIAny
AI Model2026
Icon for item

Krea 2 Identity Edit

Instruction-driven LoRA fine‑tune for identity‑preserving image edits: give an image plus a plain‑language instruction and it edits pose, outfit, objects or scene while keeping unasked content and subject likeness. Requires the ComfyUI‑Krea2Edit node pack; distributed under the Krea 2 Community License.

Introduction

Most image editors either change a person's appearance or keep identity perfectly but offer limited scene edits. This LoRA fine‑tune stitches both needs together: you give a source image and a plain‑language instruction and it produces local or global edits that preserve texture, marks and lighting cues while enabling new poses, relighting and scene swaps.

What Sets It Apart
  • Instruction‑based, identity‑focused editing: trained to follow plain English instructions (e.g., "place this person at a night market") while conserving facial texture, moles and outfit details so outputs retain recognizable likeness in most cases — useful when you need new angles or relighting without remaking the subject.
  • Locality and replace-with-reference capabilities: supports targeted add/remove/replace edits and a "replace" verb that can swap a subject with a reference object/character while keeping the rest of the frame intact, enabling compositing workflows that typical global restylers struggle with.
  • Composability with LoRAs and workflow nodes: designed to stack with other character/style LoRAs and to run in ComfyUI using the ComfyUI‑Krea2Edit node pack (dual conditioning via in‑context VAE tokens + Qwen3‑VL encoding), giving more control than closed editors.
  • Practical constraints surfaced: recommended generation at ≤2MP and matching aspect ratios to the source; two presets (Turbo for light edits, Raw for removals/large deletions) and a grounding_px dial to trade edit strength vs. likeness.
Who It's For & Trade-offs

Great fit if you need photorealistic, instructionable image edits that keep a subject's textures and small identity markers (moles, skin detail, lighting) while changing pose, outfit or scene. Works well for content creators, portfolio mockups, and iterative character restaging when you can integrate ComfyUI into your pipeline.

Look elsewhere if you require perfect preservation of unusual facial geometry (the model tends to regress extreme geometric idiosyncrasies toward typical proportions), fully reliable object removals on the first pass, or a turnkey web editor — this release is a community research/portfolio LoRA that depends on specific node support and has license conditions (Krea 2 Community License).

Where It Fits

Compared with general image editors, this model prioritizes identity fidelity over unconstrained restyling. Compared with closed commercial "face swap" tools, it offers more composability and instruction control but requires a ComfyUI‑based workflow and careful parameter tuning (grounding_px, CFG, LoRA strength) for best results.

Information

Categories

More Items

Hugging Face
AI Model2026

Post-trained Qwen3.8-27B variant using the COLD FUSION (GAIN+Unsloth) tuning to reduce internal reasoning-token use and improve instruction following while keeping base capabilities. Deliverables include 256k-context-compatible GGUF quants (regular and MTP, NEO IMATRIX), vision support via an mmproj, and three reasoning-effort modes (xhigh/medium/low).

Hugging Face
AI Model2026

Converts raw ASR transcripts into clean written text: adds punctuation and capitalization, expands spoken numbers/dates/times/currencies/emails, removes fillers and resolves self-corrections. Fine-tuned from Qwen3-0.6B (≈0.6B params), 94.8% token accuracy on a 7,519-case English test set; designed for CPU/edge deployment and deterministic post-processing.

Hugging Face
AI Model2026

GGUF build of Ornith-1.5's 35B mixture-of-experts model (A3B) for local inference — activates ~3B params per token, supports up to 262,144 tokens, emits separate reasoning traces and OpenAI-style tool calls, optimized for agentic coding and long-context use cases.