TokForge-RealisticVision-5.1-CoreML-6bit

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.

TokForge — Realistic Vision 5.1 · CoreML 6-bit (Apple Neural Engine)

A 6-bit palettized Apple CoreML conversion of Realistic Vision V5.1 (SG161222, SD-1.5 photoreal finetune), built for on-device image generation in the TokForge iOS app. Converted with Apple ml-stable-diffusion (torch2coreml) using SPLIT_EINSUM_V2 attention and --quantize-nbits 6 — 6-bit palettized weights compile fast on the Apple Neural Engine (the ANE-fast photoreal path, vs the FP16 finetunes >11-min ANE graph compile).

Part of the TokForge iOS · CoreML Image Models collection.

Files

Recommended render settings (standard SD-1.5, photoreal)

attention:    split_einsum_v2 (Apple Neural Engine)
compute:      .cpuAndNeuralEngine  (palettized -> fast ANE compile)
steps:        20   (8 = fast floor, 40 = extra refinement)
cfg-scale:    7.5
resolution:   512x512  (SD-1.5 native; baked into the compiled model)

How this was built

1. Loaded SG161222/Realistic_Vision_V5.1_noVAE (SD-1.5 diffusers; the diffusers conversion bundles a working VAE).
2. Converted UNet + text encoder + VAE decoder + VAE encoder to CoreML with Apple ml-stable-diffusion python_coreml_stable_diffusion.torch2coreml, --attention-implementation SPLIT_EINSUM_V2.
3. Applied 6-bit palettization (--quantize-nbits 6) for the fast iOS-17 ANE compile.
4. Bundled the compiled resources for the Swift CLI (--bundle-resources-for-swift-cli).

Conversion peaked at ~10.9 GB RAM (no --chunk-unet needed); iOS 17+ (6-bit palettized).

License & attribution

• 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 SG161222s.
• 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 SG161222s 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.