darkmaniac7/TokForge-SDXL-IPAdapter

SDXL IP-Adapter Plus Face (safetensors)

RealVisXL V4.0 Lightning with the SDXL IP-Adapter plus-face adapter and its image encoder, so TokForge can put one or two people from reference photos into any scene you describe.

Published by TokForge for on-device inference.

What it is	Image generator (reference photo)
Format	safetensors for stable-diffusion.cpp: f16 SDXL base in one file, plus IP-Adapter and image encoder
Runs on	Android phones with a Snapdragon chip (stable-diffusion.cpp, CPU or Adreno OpenCL GPU)
Phone class	16 GB RAM or more (TokForge catalog)
Download	10.3 GB (5 files)
Base model	SG161222/RealVisXL_V4.0_Lightning, h94/IP-Adapter
License	Several: CreativeML OpenRAIL++-M plus SG_161222's Civitai terms (base), Apache-2.0 (adapter), MIT (image encoder)

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

1. Install TokForge from Google Play.
2. On Android, open Models > Recommended > Image and download SDXL IP-Adapter (Reference Identity). TokForge lists it on phones with 16 GB RAM or more.
3. In a chat, attach a clear face photo of one person, or one photo each of two people, then describe the scene (for example "me as a superhero, unmasked, face visible"). The adapter keeps each face; the prompt drives everything else.
4. The model runs on the phone; once it is downloaded, making pictures needs no connection.

For 8 GB phones, use the lighter darkmaniac7/TokForge-SD15-IPAdapter instead.

Collections: TokForge Image Models (Android).

Files

File	Size	What it is
ip-adapter-plus-face_sdxl_vit-h.safetensors	848 MB	IP-Adapter plus-face for SDXL, ViT-H build (h94/IP-Adapter, Apache-2.0)
ip_adapter_clip_vision_vith.safetensors	2.53 GB	OpenCLIP ViT-H-14 image encoder for the reference photos (MIT)
manifest.json	1.60 KB	Render defaults, licences, sizes and md5 of the three model files
realvisxl-v40-lightning-fp16.safetensors	6.94 GB	RealVisXL V4.0 Lightning base: dual CLIP text encoders, UNet and VAE in one f16 sd.cpp safetensors
MD5SUMS	227 B	md5 of the three model files

manifest.json and MD5SUMS carry the integrity hashes and render defaults.

About the model

The highest-fidelity reference-identity image route for the TokForge Android app, and the clean multi-subject path. The plus-face IP-Adapter transfers the face only while the prompt drives the whole scene. SDXL's stronger plus-face transfer (vs SD1.5) gives sharper single-subject identity, and makes the regional-mask two-subject path produce two distinct recognizable faces.

This bundle runs on the on-device stable-diffusion.cpp engine (TokForge's IP-Adapter port) on CPU and Adreno OpenCL. Full SDXL is heavier than the SD1.5 IP-Adapter tier, so TokForge offers it on 16 GB-class phones with a Snapdragon chip.

File	License	Contents
realvisxl-v40-lightning-fp16.safetensors	OpenRAIL++	RealVisXL V4.0 Lightning (SDXL photoreal finetune), dual CLIP text encoders + UNet + VAE in one self-contained f16 sd.cpp safetensors
ip-adapter-plus-face_sdxl_vit-h.safetensors	Apache-2.0	IP-Adapter plus-face SDXL (h94/IP-Adapter), 16-token image_proj Resampler + 70 decoupled cross-attn layers (cross_attention_dim 2048)
ip_adapter_clip_vision_vith.safetensors	MIT	OpenCLIP ViT-H-14 image encoder (1280 hidden, 32 layers). The plus-face path needs ViT-H, not bigG: ViT-H is the same encoder the SD1.5 plus-face bundle ships

Why this base, and why f16

The base is RealVisXL V4.0 Lightning, a strong photoreal SDXL finetune distilled for a few-step (6-step) Lightning floor, from the same checkpoint as the TokForge RealVisXL Lightning GGUF tier. It is kept at f16 (full precision) so the IP-Adapter's decoupled cross-attention and the face Resampler keep subject quality high; a q4_0 base weakens the transferred identity (the same choice as the SD1.5 tier).

Why plus-face and ViT-H

The plus-face variant (ip-adapter-plus-face_sdxl_vit-h) runs a 16-token Resampler over the ViT-H penultimate hidden state, so it extracts the face only: identity is preserved while the prompt controls the scene. The standard ip-adapter_sdxl projects the whole pooled CLIP-bigG embedding and carries the reference's whole scene with it. Because this adapter is the _vit-h build (image_proj.latents shape [1, 16, 1280]), it pairs with the ViT-H encoder (1280 hidden), not the bigG encoder (1664 hidden) the base SDXL adapter uses. The TokForge sd.cpp IP-Adapter loader auto-detects plus-face by image_proj.latents and the SDXL adapter config (2048-dim, 70 layers) by the SDXL base.

Render settings

Setting	Value
sampler	euler
scheduler	discrete
steps	6 (Lightning few-step floor)
cfg-scale	2.0 (Lightning low-CFG)
ip-adapter-scale	0.75 (keeps the scene with strong recognizable identity; lower gives more scene freedom, higher stays closer to the reference)
resolution	1024x1024 (SDXL native)

Plus-face transfers the face only, so the rendered face must stay visible and unobstructed for a recognizable identity. Keep the face in frame ("unmasked, face visible, detailed face, looking at viewer"); the app appends this cue automatically.

TokForge runs it through its IP-Adapter port of stable-diffusion.cpp, equivalent to:

sd -M img_gen \
   -m realvisxl-v40-lightning-fp16.safetensors \
   -p "as a superhero, unmasked, face visible, detailed face" \
   -n "<strong negative>" \
   --clip_vision ip_adapter_clip_vision_vith.safetensors \
   --ip-adapter ip-adapter-plus-face_sdxl_vit-h.safetensors \
   --ip-adapter-image <your_face.jpg> \
   --ip-adapter-scale 0.75 \
   --cfg-scale 2.0 --sampling-method euler --scheduler discrete \
   --steps 6 -H 1024 -W 1024

How it was made

• Base: realvisxl-v40-lightning-fp16.safetensors, a self-contained sd.cpp SDXL safetensors built from SG161222/RealVisXL_V4.0_Lightning.
• Adapter + image encoder copied verbatim from h94/IP-Adapter (sdxl_models/ip-adapter-plus-face_sdxl_vit-h.safetensors, models/image_encoder/model.safetensors); both files are byte-identical to the upstream ones.

Limitations

• The --ip-adapter options belong to TokForge's IP-Adapter port; stock stable-diffusion.cpp builds may not have them.
• Full SDXL at 1024x1024 needs a 16 GB-class phone.

License

This is an aggregate of three independently-licensed components, and each retains its own license:

• RealVisXL V4.0 Lightning base (realvisxl-v40-lightning-fp16.safetensors): OpenRAIL++ (SG161222/RealVisXL_V4.0_Lightning, the SDXL openrail++ license). Use must comply with the OpenRAIL++ use-based restrictions.
• IP-Adapter plus-face (SDXL, ViT-H) (ip-adapter-plus-face_sdxl_vit-h.safetensors): Apache-2.0 (h94/IP-Adapter).
• OpenCLIP ViT-H-14 image encoder (ip_adapter_clip_vision_vith.safetensors): MIT (OpenCLIP / LAION ViT-H-14).

The non-commercial IP-Adapter-FaceID / InsightFace path is NOT used here, only the Apache-2.0 base + plus-face adapters from h94/IP-Adapter.

RealVisXL is also under SG_161222's Civitai terms: credit required, no selling the model or merges of it.

Credits

• RealVisXL V4.0 Lightning 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.
• Stable Diffusion XL 1.0 by Stability AI (Hugging Face). Used under the CreativeML OpenRAIL++-M license.
• IP-Adapter by Tencent AI Lab (h94) (Hugging Face). Used under the Apache-2.0 license.
• OpenCLIP ViT-H-14 image encoder by LAION (Hugging Face). Used under the MIT license.
• Runs on-device with stable-diffusion.cpp by leejet (MIT).