Realistic Vision V5.1 (Core ML 6-bit)
SG_161222's Realistic Vision V5.1 as a 6-bit Core ML bundle for making photorealistic pictures on iPhone and iPad with TokForge.
Published by TokForge for on-device inference.
| What it is | Image generator (photorealistic) |
| Format | Core ML resources for Apple's ml-stable-diffusion, 6-bit palettized, split einsum v2 attention, 512 x 512 |
| Runs on | iPhone and iPad (Core ML, Apple Neural Engine); listed in TokForge for iOS |
| Phone class | 6 GB RAM or more, iOS 17 or later (TokForge for iOS catalog) |
| Download | 1.87 GB (23 files); TokForge fetches only the zip (917 MB) |
| Base model | SG161222/Realistic_Vision_V5.1_noVAE |
| License | CreativeML OpenRAIL-M, plus SG_161222's Civitai terms (credit required, no selling) |
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 it in its image models as Realistic Vision 5.1 ยท 6-bit (ANE) and uses it for the Photoreal and Sharp (chip) looks. The app marks it as a mature (photoreal) model, so it shows up only after mature models are turned on. Pick it and the app downloads the 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 |
|---|---|---|
RealisticVision-5.1_palettized_split_einsum_v2_compiled.zip |
917 MB | Compiled Core ML resources in one zip (the file TokForge downloads) |
Resources/ |
958 MB in 22 files | The same compiled resources, unzipped: TextEncoder, Unet, VAEDecoder and VAEEncoder (.mlmodelc) plus the CLIP tokenizer files |
The zip and the Resources/ folder hold the same compiled models: .mlmodelc folders plus the CLIP vocab.json and
merges.txt, the exact layout Apple's StableDiffusionPipeline (and the TokForge installer) loads.
About the model
Realistic Vision V5.1 is SG_161222's photorealistic Stable Diffusion 1.5 fine-tune.
| Architecture | Stable Diffusion 1.5 (UNet, CLIP text encoder, VAE) |
| Weights | UNet and text encoder 6-bit palettized, VAE in fp16 |
| Components | TextEncoder, Unet, VAEDecoder and VAEEncoder (.mlmodelc), plus the CLIP vocab.json and merges.txt |
| UNet batch | 2 (classifier-free guidance works as usual) |
| Output | 512 x 512, fixed in the compiled model |
| Tooling | ml-stable-diffusion 1.1.0, coremltools 9.0, torch 2.8.0 (from each model's metadata.json) |
Render settings
Standard SD 1.5 settings for photoreal pictures:
| Setting | Value |
|---|---|
| Attention | split_einsum_v2 (Apple Neural Engine) |
| Compute units | .cpuAndNeuralEngine |
| Steps | 20 (8 is the low end, 40 adds extra refinement) |
| Guidance (CFG) | 7.5 |
| Resolution | 512 x 512 (SD 1.5 native; baked into the compiled model) |
How it was made
- Loaded
SG161222/Realistic_Vision_V5.1_noVAE(SD 1.5 diffusers; the diffusers conversion bundles a working VAE). - Converted the UNet, text encoder, VAE decoder and VAE encoder to Core ML with Apple
ml-stable-diffusionpython_coreml_stable_diffusion.torch2coreml,--attention-implementation SPLIT_EINSUM_V2. - Applied 6-bit palettization (
--quantize-nbits 6) so the Neural Engine can compile the graph without a long first-load wait. - Bundled the compiled resources for the Swift CLI (
--bundle-resources-for-swift-cli).
Conversion peaked at about 10.9 GB RAM (no --chunk-unet needed). TokForge uses the 6-bit palettized weights on
iOS 17 or later.
Limitations
- Pictures are 512 x 512 only; the size is built into the compiled model.
- 6-bit palettization may slightly reduce quality compared to the FP16 original.
- This is a Core ML runtime bundle, not a diffusers or PyTorch checkpoint.
License
Realistic Vision V5.1 and this conversion are under the CreativeML OpenRAIL-M license, and SG_161222's Civitai terms apply too: credit required, no selling the model or merges of it, and derivatives keep the same permissions (see Credits).
- License: CreativeML OpenRAIL-M (inherited from Realistic Vision V5.1 / Stable Diffusion 1.5). Subject to OpenRAIL-M restrictions.
- Base model: Realistic Vision V5.1 by SG161222: https://huggingface.co/SG161222/Realistic_Vision_V5.1_noVAE (an SD-1.5 finetune). All credit for the model weights is SG161222's.
- Conversion tooling: Apple
ml-stable-diffusion: https://github.com/apple/ml-stable-diffusion (6-bit palettization,SPLIT_EINSUM_V2). - Built on top of Stable Diffusion 1.5 (Runway/CompVis/Stability).
This repository is a redistribution for on-device use, a format conversion (PyTorch -> CoreML) and 6-bit palettization of SG161222's Realistic Vision V5.1. No weights were retrained. The original OpenRAIL-M terms and attribution requirements propagate to this conversion and any images generated with it. No additional restrictions are imposed by this repackaging.
Credits
- Realistic Vision V5.1 by SG_161222 (Civitai, Hugging Face). Used under the CreativeML OpenRAIL-M license and the creator's Civitai terms. SG_161222's Civitai terms: credit required, no selling the model or merges of it, and derivatives keep the same permissions.
- 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).
Model tree for darkmaniac7/TokForge-RealisticVision-5.1-CoreML-6bit
Base model
SG161222/Realistic_Vision_V5.1_noVAE