ReV Animated and SDXL-Turbo (Core ML 6-bit)
Two image models for TokForge on iPhone and iPad, each converted to 6-bit Core ML: ReV Animated and SDXL-Turbo.
Published by TokForge for on-device inference.
| What it is | Image generator (two models in one repo, one folder each) |
| Format | Core ML resources for Apple's ml-stable-diffusion, 6-bit palettized, split einsum attention, 512 x 512 |
| Runs on | iPhone and iPad (Core ML, Apple Neural Engine); listed in TokForge for iOS |
| Phone class | ReV Animated: 4 GB RAM or more. SDXL-Turbo: 12 GB RAM or more. Both iOS 17 or later (TokForge for iOS catalog) |
| Download | 10.2 GB (56 files); TokForge fetches only one zip per model (854 MB for ReV Animated, 2.74 GB for SDXL-Turbo) |
| Base model | stabilityai/sdxl-turbo, ReV Animated by Zovya (Civitai) |
| License | Two licences: CreativeML OpenRAIL-M plus Zovya's Civitai terms (ReV Animated); Stability AI Community License (SDXL-Turbo) |
TokForge
- Website: https://tokforge.ai
- Discord: https://discord.gg/Acv3CBtfVm
- Google Play: https://play.google.com/store/apps/details?id=dev.tokforge
- iOS TestFlight: https://testflight.apple.com/join/jnufjzRr
Runs on-device in the TokForge app.
Use it in TokForge
- Join the iOS TestFlight to install TokForge on iPhone or iPad.
- TokForge for iOS lists both in its image models: ReV Animated · 6-bit (ANE) for semi-realistic, fantasy and stylized art, and SDXL-Turbo · 6-bit (ANE) for few-step pictures on devices with 12 GB RAM or more. Pick one and the app downloads that model's zip, unpacks it and runs it on the Neural Engine.
- Describe the picture you want and generate it. The model runs on the device; once it is downloaded, making pictures needs no connection.
Collections: TokForge iOS Models.
Files
| File | Size | What it is |
|---|---|---|
TokForge-ReV-Animated-CoreML-6bit/ |
1.74 GB in 18 files | ReV Animated, 6-bit: the compiled zip TokForge downloads plus the same Resources/ tree unzipped |
TokForge-SDXL-Turbo-CoreML-6bit/ |
8.49 GB in 38 files | SDXL-Turbo, 6-bit: the compiled zip TokForge downloads, the same Resources/ tree unzipped, the uncompiled .mlpackage sources, a README, the licence text and the notice |
Each folder has its own compiled zip (the file TokForge downloads) and the same Resources/ tree unzipped. The
SDXL-Turbo folder also keeps the uncompiled .mlpackage sources, its own
README with the full conversion details, the licence text
(LICENSE.md) and the required notice (NOTICE).
What is in each folder
| Folder | Model | Base | Components | TokForge settings |
|---|---|---|---|---|
TokForge-ReV-Animated-CoreML-6bit/ |
ReV Animated, 6-bit | Stable Diffusion 1.5 fine-tune by Zovya | TextEncoder, Unet, VAEDecoder (no VAE encoder, so text to image only) |
20 steps (8 to 40), guidance 7.0 |
TokForge-SDXL-Turbo-CoreML-6bit/ |
SDXL-Turbo, 6-bit | stabilityai/sdxl-turbo (adversarially distilled SDXL) | TextEncoder, TextEncoder2, Unet, VAEDecoder |
2 steps (1 to 4), guidance 1.0 |
Both UNets are baked for a batch of two, the layout Apple's pipelines feed on every step. Both output 512 x 512
pictures; the size is built into the compiled models. ReV Animated runs on Apple's StableDiffusionPipeline;
SDXL-Turbo needs StableDiffusionXLPipeline (iOS 17 or later). SDXL-Turbo was trained without classifier-free
guidance, so guidance 1.0 is its guidance-free setting on Apple's pipeline.
How it was made
Both models were converted with Apple
ml-stable-diffusion 1.1.0 and coremltools 9.0 (per each model's
metadata.json), with 6-bit palettization (--quantize-nbits 6):
- ReV Animated:
SPLIT_EINSUM_V2attention; the UNet and text encoder are 6-bit palettized, the VAE decoder is fp16. - SDXL-Turbo:
SPLIT_EINSUMattention,--latent-h 64 --latent-w 64; the UNet and both text encoders are 6-bit palettized, the VAE decoder is float32. The weights were not retrained or fine-tuned.
Limitations
- Pictures are 512 x 512 only.
- SDXL-Turbo needs iOS 17 or later and a device with 12 GB RAM or more in TokForge.
- ReV Animated has no VAE encoder, so it does text to image only.
- These are Core ML runtime bundles, not diffusers or PyTorch checkpoints.
License
This repo holds two models under different licenses:
TokForge-ReV-Animated-CoreML-6bit/: the CreativeML OpenRAIL-M license plus Zovya's Civitai terms for ReV Animated (credit required, no selling the model or merges of it, no use on other image generation services).TokForge-SDXL-Turbo-CoreML-6bit/: the Stability AI Community License by Stability AI. Free commercial use is limited to individuals and organizations with under USD 1M in annual revenue, commercial use requires registration with Stability AI, and the license's notice terms apply to this folder: "This Stability AI Model is licensed under the Stability AI Community License, Copyright © Stability AI Ltd. All Rights Reserved." and "Powered by Stability AI".
The repo metadata declares the more restrictive of the two, the Stability AI Community License.
The SDXL-Turbo folder carries a copy of the Stability AI Community License in LICENSE.md and the required notice in
NOTICE.
Credits
- ReV Animated by Zovya (Civitai). Used under the CreativeML OpenRAIL-M license and the creator's Civitai terms. Zovya's Civitai terms: credit required, no selling the model or merges of it, no use on other image generation services, and derivatives keep the same permissions. Contained in
TokForge-ReV-Animated-CoreML-6bit/. - SDXL-Turbo by Stability AI (Hugging Face). Used under the Stability AI Community License. Contained in
TokForge-SDXL-Turbo-CoreML-6bit/. This Stability AI Model is licensed under the Stability AI Community License, Copyright © Stability AI Ltd. All Rights Reserved. Powered by Stability AI. - Stable Diffusion v1.5 by Runway and CompVis (Hugging Face). Used under the CreativeML OpenRAIL-M license.
- Converted to Core ML with ml-stable-diffusion by Apple (MIT).
- Downloads last month
- 25
Model tree for darkmaniac7/tokforge-ios-coreml-image-models
Base model
stabilityai/sdxl-turbo