Vathography/marigold-v2-see-through-comfyui

Marigold V2 see-through depth for ComfyUI

The see-through depth model of Marigold V2 (depth/Log-layered): it estimates the depth behind glass instead of stopping at the glass. Converted for ComfyUI, on the 4-bit (nf4) Qwen-Image-Edit-2509 backbone it was trained with. Independent conversion, not affiliated with Marigold V2 or ComfyUI.

Files

File	Destination folder
marigold_v2_see_through_qwen_image_edit_2509_nf4.safetensors	models/diffusion_models/
marigold_v2_depth_see_through.safetensors	models/loras/
marigold_v2_depth_see_through_vae.safetensors	models/vae/

Also needed: embeddings/marigold_v2_depth_conditioning.safetensors from Comfy-Org/marigold-v2-0.

How to use in ComfyUI

With the see-through example workflow of the Marigold V2 Tiled Depth - Vathography node pack (its Marigold V2 See-through Loader loads these files). Estimated minimum: 8 GB VRAM + 24 GB system RAM.

Changes made

Conversion only, no retraining: the backbone quantised to nf4 (bitsandbytes, blocksize 64; lossy, as in the original training setup), the LoRA renamed to ComfyUI's keys, the fine-tuned VAE decoder placed into the full VAE with the unchanged Qwen-Image-Edit-2509 encoder. Details in NOTICE.

Credits and license

Marigold V2 — Igor Pavlovic, Thiemo Wandel, Anton Obukhov, Luca Bartolomei, Andrey Davydov, Fabio Tosi, Matteo Poggi, Sabine Süsstrunk, Dengxin Dai (EPFL, HUAWEI Bayer Lab, University of Bologna). Paper · code · weights. Qwen-Image-Edit-2509 — the Qwen team. Model · report. ComfyUI Marigold V2 support and file layout — Comfy-Org.

Apache License 2.0 (LICENSE), as the original works. Free to use, including commercially. An independent conversion, not affiliated with or endorsed by the authors above.

@article{pavlovic2026marigoldv2,
    author = {Pavlovic, Igor and Wandel, Thiemo and Obukhov, Anton and Bartolomei, Luca and Davydov, Andrey and Tosi, Fabio and Poggi, Matteo and S{\"u}sstrunk, Sabine and Dai, Dengxin},
    title = {Marigold V2: Revisiting Diffusion Transformers for Monocular Depth Estimation},
    journal = {ACM Trans. Graph.}, volume = {45}, number = {6}, articleno = {204}, year = {2026},
    doi = {10.1145/3842528}
}
@misc{wu2025qwenimagetechnicalreport,
    title = {Qwen-Image Technical Report},
    author = {Chenfei Wu and others},
    year = {2025}, eprint = {2508.02324}, archivePrefix = {arXiv}
}