fix edit provenance

README.md

@@ -111,5 +111,7 @@ These blocks only work with Krea 2, Turbo or raw.
   style-reference LoRA behind `append`
 - [`conradlocke/krea2-identity-edit`](https://huggingface.co/conradlocke/krea2-identity-edit) -- the
   Identity-Edit LoRA behind `prepend`
-- Both recipes come from [ai-toolkit](https://github.com/ostris/ai-toolkit); `append` follows its
-  reference/edit setup, `prepend` its `predict_velocity_edit`.
+- `append` follows the built-in Krea 2 edit arch in [ai-toolkit](https://github.com/ostris/ai-toolkit);
+  `prepend` follows `predict_velocity_edit` in
+  [krea2edit-trainer](https://github.com/lbouaraba/krea2edit-trainer), the third-party ai-toolkit extension
+  by Conrad Locke the Identity-Edit LoRA was trained with

krea2_reference.py

@@ -22,7 +22,7 @@ latents* inside the transformer sequence. All three paths live here, selected by
   time `t=0`. The Ostris AI-Toolkit reference/edit LoRA recipe.
 * `"prepend"` -- vision (grounded view) + the clean VAE source **before** the noisy target under the plain
   uniform timestep modulation, distinguished only by its RoPE frame axis. The Identity-Edit recipe
-  (ai-toolkit `predict_velocity_edit`); its unconditional branch is grounded on the same image.
+  (krea2edit-trainer's `predict_velocity_edit`); its unconditional branch is grounded on the same image.
 
 With `reference_mode="off"` and no images the blocks reduce to stock text-to-image. No mode patches the
 `Krea2Transformer2DModel` source.