Skip to content
Images

DESPINA lunar SDXL: Lunar surface views generation

2 min readupdated

This release is a set of trained weights based on Stable Diffusion XL designed to turn digital elevation models (DEMs) into realistic photographs of the lunar horizon. The system uses two ControlNet modules for depth and soft-edge analysis, along with a special LoRA add-on to reproduce an authentic lunar look based on Apollo archive imagery. The resulting horizon and topography strictly match the original geographic data. The tool is geared toward space researchers, planetary scientists, and visualization developers.

Key features

  • Precise terrain anchoring: Uses an inverted depth map and contours to control generation pixel by pixel.
  • Lunar sky: A specially trained LoRA supports rendering a black sky with correct noise.
  • Comprehensive approach: Includes three components—a lunar LoRA for UNet, depth ControlNet, and soft-edge ControlNet.
  • Scientific base: Built on academic research (ISPRS Annals) using ASU Apollo scans.

What you can do

  • Landscape reconstruction: Create photorealistic panoramas of the lunar surface based on real elevation data (DEM).
  • Batch rendering: Use the project's repository scripts to generate lunar views for specified camera coordinates and sun angles.
  • Diffusers integration: Run generation directly via the Stable Diffusion XL ControlNet pipeline in Python scripts.

Example

import torch
from diffusers import ControlNetModel, StableDiffusionXLControlNetPipeline

repo = "RonInDune/despina-lunar-sdxl"
cns = [ControlNetModel.from_pretrained(repo, subfolder="controlnet-depth"),
       ControlNetModel.from_pretrained(repo, subfolder="controlnet-softedge")]
pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0", controlnet=cns).to("cuda")
pipe.load_lora_weights(repo, subfolder="lora", weight_name="pytorch_lora_weights.safetensors")
pipe.fuse_lora(lora_scale=0.9)

image = pipe(
    "A realistic image of the Moon's surface, from the perspective of a rover. Black sky.",
    image=[depth_image, softedge_image],
    controlnet_conditioning_scale=[0.8, 0.5],
    num_inference_steps=30, guidance_scale=7.0,
).images[0]
Read moreDetails
• Accuracy: average horizon error Horizon RMSE 0.94 px at 512 px.
• Training: 3000 steps on 1277 frames in fp32 using diffusers 0.31.0.
• Limitations: non-commercial use (research and education) due to base model and NASA/JSC/ASU data licenses.
Discuss Telegram 0

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 14 GB · ready files17 GB
  • BF16weights 9.4 GB12 GB
  • FP8 / INT8recommendedweights 7.1 GB9.3 GB
  • 4-bitweights 6.0 GB8.1 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 3080 · F32 — from RTX 4090

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model stable-diffusion-xl-base-1.0.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # RonInDune/despina-lunar-sdxl
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download stabilityai/stable-diffusion-xl-base-1.0 --local-dir ~/models/stable-diffusion-xl-base-1.0
    hf download RonInDune/despina-lunar-sdxl --local-dir ~/models/despina-lunar-sdxl
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • SageAttention 3v2.2.0from October 28, 2025

    even faster on RTX 50 / Blackwell

    FP4 attention for Blackwell. Faster than SageAttention 2, but the authors suggest v2 when accuracy matters.

    How to enable: Build from the sageattention3_blackwell folder of the repo; in ComfyUI via KJNodes.

    Earlier generations (2)
    • SageAttention 2++ — Quantised (INT8/FP8) attention with no visible quality loss. The main speed-up for image and video generation.
    • SageAttention 1 — First version: INT8 for Q·K. Still works, but 2++ is faster and more accurate.
  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (3)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
    • xFormers — The older way to speed up attention, works almost everywhere. Sage or Flash are faster where available.
  • cache-dit (DBCache)v1.5.6from November 20, 2026new release

    1.5–3× at nearly the same quality

    Reuses block outputs between diffusion steps. Supports most DiT models.

    How to enable: pip install cache-dit; one line cache_dit.enable_cache(pipe) for diffusers.

    Earlier generations (2)
    • MagCache — Skips steps by magnitude of change. More accurate than TeaCache, no calibration.
    • TeaCache — The first popular diffusion step cache. Works, newer methods are more accurate.
  • Nunchaku (SVDQuant 4-bit)v1.2.1from January 25, 2026

    ≈3× faster and 3.5× less memory

    4-bit versions of popular image models with almost no quality loss (FP4 on RTX 50).

    How to enable: Ready weights for supported models; in ComfyUI use ComfyUI-nunchaku.

  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

Requires the base model

stabilityai/stable-diffusion-xl-base-1.0

This model is a LoRA / adapter. Download the base model first, then load this one on top.

New posts about versions of stable-diffusion-xl-base-1.0 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 5All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face
Merges 2All on Hugging Face

About the model

RonInDune/despina-lunar-sdxl
Author
RonInDune
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments