Third-Party Model Notices
Effective: September 1, 2026
1. Scene Continuity Anime Image Model
For the Scene Continuity feature, Lucidream uses Animagine XL 4.0 Opt, developed by Cagliostro Research Lab and fine-tuned from Stable Diffusion XL 1.0, to generate anime scene images.
Exact artifact: cagliostrolab/animagine-xl-4.0 at revision 2b7c1b397761bf5bd3cc42e5b39ec99314a75a96; file animagine-xl-4.0-opt.safetensors; 6,938,350,040 bytes; SHA-256 6327eca98bfb6538dd7a4edce22484a1bbc57a8cff6b11d075d40da1afb847ac. The corresponding Civitai record is model 1188071, version 1408658.
This entry covers the Scene Continuity anime image model only; it is not a complete inventory of every AI model used elsewhere in the Service.
2. Reference and Pose Conditioning Models
IP-Adapter Plus SDXL ViT-H — h94/IP-Adapter at revision 018e402774aeeddd60609b4ecdb7e298259dc729; file sdxl_models/ip-adapter-plus_sdxl_vit-h.safetensors; 847,517,512 bytes; SHA-256 3f5062b8400c94b7159665b21ba5c62acdcd7682262743d7f2aefedef00e6581. The pinned Hugging Face repository card declares Apache-2.0. Source attribution: h94/IP-Adapter contributors.
IP-Adapter Plus Face SDXL ViT-H — sdxl_models/ip-adapter-plus-face_sdxl_vit-h.safetensors from the same h94/IP-Adapter revision; 847,517,512 bytes; SHA-256 677ad8860204f7d0bfba12d29e6c31ded9beefdf3e4bbd102518357d31a292c1. The Step 3 Animagine reference/OpenPose workflow selects this non-FaceID weight at scale 0.50. It applies CLIP image-encoder conditioning to a deterministic system-anchor head-and-torso crop and does not perform biometric face recognition. The pinned repository card declares Apache-2.0; source attribution: h94/IP-Adapter contributors.
CLIP Vision ViT-H image encoder — models/image_encoder/model.safetensors from the same h94/IP-Adapter revision; 2,528,373,448 bytes; SHA-256 6ca9667da1ca9e0b0f75e46bb030f7e011f44f86cbfb8d5a36590fcd7507b030. It is packaged with the IP-Adapter snapshot; we preserve the pinned repository card's Apache-2.0 declaration and the upstream OpenCLIP/LAION provenance.
OpenPose SDXL ControlNet — xinsir/controlnet-openpose-sdxl-1.0 at revision 23f966cd5cfdd3f7729c903e243d87152162d2b7; file diffusion_pytorch_model.safetensors; 2,502,139,104 bytes; SHA-256 b8524e557a7df60d081f5d4a0eb109967d107df217943bf88c2d99b9ebcc06c5. The pinned Hugging Face repository card declares Apache-2.0. Source attribution: xinsir.
The two pinned model snapshots do not contain a separate LICENSE file, so Lucidream preserves the Apache License 2.0 text together with each pinned repository README and model card. No FaceID, InsightFace, InstantID, or PuLID asset is included in this configuration.
3. License and Attribution
The Animagine XL 4.0 model card states that the model adopts the original CreativeML Open RAIL++-M License from Stability AI without modifications or additional restrictions. License notice: Copyright (c) 2023 Stability AI; CreativeML Open RAIL++-M License dated July 26, 2023. Animagine XL 4.0 attribution: Cagliostro Research Lab.
The license permits hosted remote access subject to its conditions and Attachment A use restrictions. Those restrictions are included as enforceable provisions in Section 5 of the Lucidream Terms of Service and apply to every user of the relevant model-backed features.
4. Lucidream Service-Layer Changes
Lucidream uses the published checkpoint bytes identified above without modifying that checkpoint file. Around the checkpoint, Lucidream applies its own inference workflow, prompts, safety filtering, output scanning, and reference-image conditioning. These are service-layer changes, and Lucidream does not offer the checkpoint itself for download.
If the active model artifact or these material service-layer changes are replaced, this notice will be updated with the new identity and change statement.
5. Outputs and Third-Party Rights
Data, as defined by the model license, is not licensed under that license. The Model and Complementary Material are provided on an AS-IS basis without warranties of title, non-infringement, merchantability, or fitness for a particular purpose.
The license states that, except as set out there, its licensor claims no rights in model-generated Output, while making users accountable for outputs and prohibiting output uses that contravene the license. That provision does not clear third-party rights, change the Lucidream Terms, or grant or promise broader output rights.
Users remain responsible for their prompts, outputs, and subsequent uses, including rights of publicity, privacy, copyright, trademark, and other applicable rights and laws.