v2.6

• WebDSText2ImageSource: images in webdataset and text captions in label_file (.txt, .json, ...)
• Add drop probability option to TagDropoutHandler
• Fix GWLoss
• Remake sampler
• Add solver to sampler (Eular ODE solver, DPM dolver, ...)
• Fix bugs of attention_mask in TokenizeHandler
• Preview images with logger (preview to wandb/tensorboard, supported by rainbowneko>=1.10)
• Cache trainable_parameters in wrapper to reduce overhead. (Iteration over trainable_parameters is time-consuming)
• Support VAEs with latents_mean and latents_std

v2.4 Flux Support In-Progess

• Fix T5Encoder for pixart
• Add PixartDenoiseAction
• Add CharbonnierLoss
• Fix add_noise in DiffusersSampler

Flux Support In-Progess

We start to support flow matching models (e.g. FLUX).

• Add FluxShiftTimeSampler
• Remake ComposeTextEncoder to better support flux and other models.

  Change from return a composed embedding to return a dict of embeddings.

• Support input with more shapes (B,C,H,W) and (B,L,C)
• Flux vae support

v2.3.1

Add evaluation example configs.

v2.3

Evaluation

Add evaluators for select best ckpt and help for adjust hyperparameters. Support CLIPScore, FID, CCIPScore, ...

Sampler

Remake samplers using a new paradigm:
$$x_t = \alpha_t x_0 + \sigma_t \epsilon$$
$$x_0 = c_{skip} x_t + c_{out} F_{\theta} \left(c_{in} x_t, c_{noise}(t)\right)$$
$$v_t = \frac{dx_t}{dt}$$

• Support target type convert ($\epsilon \rightarrow x_0$, $\epsilon \rightarrow v_t$, ...)
• Support VP-SDE (DDPM, VP, ...)
• Support flow matching
• Support EDM

Updates:

• Fix file name bugs in SaveImageAction.
• Update to rainbowneko==1.9.
• Support save optimizer states.
• Support decompile config object into .py file.

v2.2.1

• deepspeed fix
• update to rainbowneko==1.6

v2.2

• Lora support webui format
• Update documents
• Workflow support generate images from text dataset

HCP-Diffusion V2

HCP-Diffusion V2 is built on top of the 🐱 RainbowNeko Engine. Using python format config file instead of yaml, which can be more extensible, flexible, and user-friendly.

• More user-friendly config file
• Simplified configuration
• More user-friendly model evaluation
• More model and method support

✨ Features

📦 Model Support

Model Name	Status
Stable Diffusion 1.5	✅ Supported
Stable Diffusion XL (SDXL)	✅ Supported
PixArt	✅ Supported
FLUX	🚧 In Development
Stable Diffusion 3 (SD3)	🚧 In Development

---

🧠 Fine-Tuning Capabilities

Feature	Description/Support
LoRA Layer-wise Configuration	✅ Supported (including Conv2d)
Layer-wise Fine-Tuning	✅ Supported
Multi-token Prompt-Tuning	✅ Supported
Layer-wise Model Merging	✅ Supported
Custom Optimizers	✅ Supported (Lion, DAdaptation, pytorch-optimizer, etc.)
Custom LR Schedulers	✅ Supported

---

🧩 Extension Method Support

Method	Status
ControlNet (including training)	✅ Supported
DreamArtist / DreamArtist++	✅ Supported
Token Attention Adjustment	✅ Supported
Max Sentence Length Extension	✅ Supported
Textual Inversion (Custom Tokens)	✅ Supported
CLIP Skip	✅ Supported

---

🚀 Training Acceleration

Tool/Library	Supported Modules
🤗 Accelerate	✅ Supported
Colossal-AI	✅ Supported
xFormers	✅ Supported (UNet and text encoder)

---

🗂 Dataset Support

Feature	Description
Aspect Ratio Bucket (ARB)	✅ Auto-clustering supported
Multi-source / Multi-dataset	✅ Supported
LMDB	✅ Supported
webdataset	🚧 In Development
Local Attention Enhancement	✅ Supported
Tag Shuffling & Dropout	✅ Multiple tag editing strategies

---

📉 Supported Loss Functions

Loss Type	Description
Min-SNR	✅ Supported
SSIM	✅ Supported
GWLoss	✅ Supported

---

🌫 Supported Diffusion Strategies

Strategy Type	Status
DDPM	✅ Supported
EDM	✅ Supported
Flow Matching	✅ Supported

---

🧠 Automatic Evaluation (Step Selection Assistant)

Feature	Description/Status
Image Preview	✅ Supported (workflow preview)
FID	🚧 In Development
CLIP Score	🚧 In Development
CCIP Score	🚧 In Development
Corrupt Score	🚧 In Development

v0.9.1

doc controlnet

V0.9.0

• Fix file extension bug in CaptionLoader
• Fix WrapablePlugin
• Support cache RatioBucket
• Multi-threaded bucket building
• Add TagErase for caption porcess
• Add workflow features.
• Fix controlnet conversion
• Support plugin with partial trainable parameters
• Refactoring lora to improve training speed
• support encoder_attention_mask and position ids
• Add SoftMinSNRLoss
• Fix model EMA

v0.7.0

• Separate data_source into a separate class that can be modified in the configuration file. The kohya datasource format can be supported.
• Fix bugs in preview images
• Fix bugs in warp_models
• Rename tag_transforms into text_transforms