Instructions to use jejunepixels/Hammershoi-Ideogram4-LoRA with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use jejunepixels/Hammershoi-Ideogram4-LoRA with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("ideogram-ai/ideogram-4-fp8", dtype=torch.bfloat16, device_map="cuda") pipe.load_lora_weights("jejunepixels/Hammershoi-Ideogram4-LoRA") prompt = "{\"high_level_description\":\"hmrsh painting: a wide view of a woman in a dark dress seen from behind, an open doorway leading to a sequence of rooms\",\"style_description\":{\"aesthetics\":\"quiet, austere, contemplative, still\",\"lighting\":\"soft diffused light\",\"medium\":\"painting\",\"art_style\":\"oil on canvas, danish symbolism, muted palette\",\"color_palette\":[\"#A39A88\",\"#2A2622\",\"#D8D2C4\",\"#6F685A\"]},\"compositional_deconstruction\":{\"background\":\"pale panelled walls and dark wooden floors\",\"elements\":[{\"type\":\"obj\",\"bbox\":[300,620,950,860],\"desc\":\"woman in a dark dress seen from behind, hair pinned up\"},{\"type\":\"obj\",\"bbox\":[180,180,800,560],\"desc\":\"open doorway to a sequence of rooms with a distant window\"}]}}" image = pipe(prompt).images[0] - Inference
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
- DiffusionBee
import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("ideogram-ai/ideogram-4-fp8", dtype=torch.bfloat16, device_map="cuda")
pipe.load_lora_weights("jejunepixels/Hammershoi-Ideogram4-LoRA")
prompt = "{\"high_level_description\":\"hmrsh painting: a wide view of a woman in a dark dress seen from behind, an open doorway leading to a sequence of rooms\",\"style_description\":{\"aesthetics\":\"quiet, austere, contemplative, still\",\"lighting\":\"soft diffused light\",\"medium\":\"painting\",\"art_style\":\"oil on canvas, danish symbolism, muted palette\",\"color_palette\":[\"#A39A88\",\"#2A2622\",\"#D8D2C4\",\"#6F685A\"]},\"compositional_deconstruction\":{\"background\":\"pale panelled walls and dark wooden floors\",\"elements\":[{\"type\":\"obj\",\"bbox\":[300,620,950,860],\"desc\":\"woman in a dark dress seen from behind, hair pinned up\"},{\"type\":\"obj\",\"bbox\":[180,180,800,560],\"desc\":\"open doorway to a sequence of rooms with a distant window\"}]}}"
image = pipe(prompt).images[0]Hammershoi Ideogram 4 LoRA

- Prompt
- {"high_level_description":"hmrsh painting: a wide view of a woman in a dark dress seen from behind, an open doorway leading to a sequence of rooms","style_description":{"aesthetics":"quiet, austere, contemplative, still","lighting":"soft diffused light","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#A39A88","#2A2622","#D8D2C4","#6F685A"]},"compositional_deconstruction":{"background":"pale panelled walls and dark wooden floors","elements":[{"type":"obj","bbox":[300,620,950,860],"desc":"woman in a dark dress seen from behind, hair pinned up"},{"type":"obj","bbox":[180,180,800,560],"desc":"open doorway to a sequence of rooms with a distant window"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a close-up of a young woman with dark hair pinned up, in three-quarter view with a downcast gaze","style_description":{"aesthetics":"quiet, luminous, austere, still","lighting":"bright diffused light","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#D8D2C4","#A39A88","#2A2622","#6F685A"]},"compositional_deconstruction":{"background":"pale neutral background","elements":[{"type":"obj","bbox":[180,280,760,740],"desc":"face in three-quarter view with downcast eyes"},{"type":"obj","bbox":[60,240,330,780],"desc":"dark hair pinned up"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a wide view of a dark room with a round table holding two lit candles in tall candlesticks","style_description":{"aesthetics":"quiet, intimate, hushed, still","lighting":"dark interior, candlelight","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#242422","#3A3734","#5C5146","#C9A86A"]},"compositional_deconstruction":{"background":"dark room in deep shadow","elements":[{"type":"obj","bbox":[260,420,620,580],"desc":"two lit candles in tall silver candlesticks"},{"type":"obj","bbox":[580,200,950,800],"desc":"round table with a pale cloth catching the candlelight"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a wide view of a woman in a dark dress silhouetted at a tall window with white curtains","style_description":{"aesthetics":"quiet, introspective, luminous, still","lighting":"backlight from window","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#2A2622","#D8D2C4","#6F685A","#A39A88"]},"compositional_deconstruction":{"background":"dark walls in a sparse room","elements":[{"type":"obj","bbox":[340,380,960,620],"desc":"woman in a dark dress seen from behind"},{"type":"obj","bbox":[60,220,720,780],"desc":"tall window with white curtains parted"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a wide view of an empty room with a tall window casting a grid of light on the wooden floor","style_description":{"aesthetics":"quiet, austere, contemplative, still","lighting":"soft diffused light","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#A39A88","#2A2622","#D8D2C4","#6F685A"]},"compositional_deconstruction":{"background":"empty room with pale walls","elements":[{"type":"obj","bbox":[80,340,620,680],"desc":"tall multi-paned window"},{"type":"obj","bbox":[720,280,960,760],"desc":"grid of window light on a dark wooden floor"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a wide view of dark sailing ships moored in a harbour with crossing masts and rigging","style_description":{"aesthetics":"quiet, misty, melancholy, still","lighting":"overcast daylight","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#A8A296","#2F2C28","#7A5A3C","#8E8779"]},"compositional_deconstruction":{"background":"pale grey overcast sky above grey water","elements":[{"type":"obj","bbox":[100,60,820,700],"desc":"dark sailing ships with crossing masts and rigging"},{"type":"obj","bbox":[800,0,1000,1000],"desc":"grey water"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a close-up of a closed silver laptop and a white ceramic mug on a dark wooden desk","style_description":{"aesthetics":"quiet, austere, contemplative, still","lighting":"soft diffused light","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#A39A88","#2A2622","#D8D2C4","#6F685A"]},"compositional_deconstruction":{"background":"pale wall behind, diffused window light","elements":[{"type":"obj","bbox":[520,120,780,640],"desc":"closed silver laptop"},{"type":"obj","bbox":[400,680,760,900],"desc":"white ceramic mug with black coffee"},{"type":"obj","bbox":[700,0,1000,1000],"desc":"dark wooden desk"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a wide view of two women in dark clothing at a table by a window in a sparse room","style_description":{"aesthetics":"quiet, austere, contemplative, still","lighting":"soft diffused light","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#A39A88","#2A2622","#D8D2C4","#6F685A"]},"compositional_deconstruction":{"background":"pale walls and a wooden floor","elements":[{"type":"obj","desc":"two women in dark clothing seated"},{"type":"obj","desc":"small table with a white cloth"},{"type":"obj","desc":"tall window with white curtains"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a woman in a dark dress seen from behind in a sparse room","style_description":{"aesthetics":"quiet, austere, contemplative, still","lighting":"soft diffused light","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#A39A88","#2A2622","#D8D2C4","#6F685A"]},"compositional_deconstruction":{"background":"sparse room with pale walls and a grey wooden floor","elements":[{"type":"obj","bbox":[250,80,950,380],"desc":"woman in a dark dress seen from behind, hair pinned up"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a sparse room with a dark wooden chair and a small round table","style_description":{"aesthetics":"quiet, austere, contemplative, still","lighting":"soft diffused light","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#A39A88","#2A2622","#D8D2C4","#6F685A"]},"compositional_deconstruction":{"background":"sparse room with pale walls and a grey wooden floor","elements":[{"type":"obj","bbox":[450,80,950,400],"desc":"dark wooden chair"},{"type":"obj","bbox":[500,550,950,950],"desc":"small round table with a white cloth"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a woman in a dark dress seen from behind in a sparse room","style_description":{"aesthetics":"quiet, austere, contemplative, still","lighting":"soft diffused light","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#A39A88","#2A2622","#D8D2C4","#6F685A"]},"compositional_deconstruction":{"background":"sparse room with pale walls and a grey wooden floor","elements":[{"type":"obj","bbox":[600,440,880,560],"desc":"woman in a dark dress seen from behind, hair pinned up"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a sparse room with a window, a door, an iron stove and a small framed picture","style_description":{"aesthetics":"quiet, austere, contemplative, still","lighting":"soft diffused light","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#A39A88","#2A2622","#D8D2C4","#6F685A"]},"compositional_deconstruction":{"background":"sparse room with pale walls and a grey wooden floor","elements":[{"type":"obj","bbox":[80,60,600,340],"desc":"tall multi-paned window"},{"type":"obj","bbox":[120,400,300,620],"desc":"small framed picture on the wall"},{"type":"obj","bbox":[380,380,880,600],"desc":"large dark iron stove"},{"type":"obj","bbox":[180,680,900,940],"desc":"white panelled door"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a landscape with a ploughed field and a row of bare trees under an overcast sky","style_description":{"aesthetics":"quiet, open, melancholy, still","lighting":"overcast daylight","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#B5B2A6","#7A5A3C","#5C6448","#2F2A24"]},"compositional_deconstruction":{"background":"open countryside","elements":[{"type":"obj","bbox":[0,0,780,1000],"desc":"pale grey overcast sky"},{"type":"obj","bbox":[690,100,790,800],"desc":"row of bare autumn trees"},{"type":"obj","bbox":[780,0,1000,1000],"desc":"wide orange ploughed field"}]}}

- Prompt
- {"high_level_description":"hmrsh painting: a landscape with a ploughed field and a row of bare trees under an overcast sky","style_description":{"aesthetics":"quiet, open, melancholy, still","lighting":"overcast daylight","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#B5B2A6","#7A5A3C","#5C6448","#2F2A24"]},"compositional_deconstruction":{"background":"open countryside","elements":[{"type":"obj","bbox":[0,0,220,1000],"desc":"pale grey overcast sky"},{"type":"obj","bbox":[140,100,230,800],"desc":"row of bare autumn trees"},{"type":"obj","bbox":[220,0,1000,1000],"desc":"wide orange ploughed field"}]}}
Model description
A LoRA fine-tuned on Ideogram 4 to generate images in the quiet, introspective style of Danish painter Vilhelm Hammershøi (1864–1916). Characterized by muted grey palettes, sparse domestic interiors, solitary figures, and profound stillness.
Unlike the FLUX versions, this LoRA was trained on Ideogram 4's structured JSON captions, including bounding boxes, so you can place objects in the image. See also the Flux1 version and the Flux2 version.
It is intended for non-commercial art and style exploration.
Demo
Try this model in your browser, with and without the LoRA, and with bounding boxes: Hammershoi Ideogram 4 LoRA Demo. You sign in with your Hugging Face account, and generations run on your own Inference Providers quota.
Trigger words
You should use hmrsh to trigger the image generation. Start high_level_description with hmrsh painting:.
Recommended settings
| Parameter | Value |
|---|---|
| Steps | 30 |
| Guidance | 4.0 |
| LoRA scale | 1.0 |
| Resolution | 1024×1024 (square) |
| Sampler | Flowmatch |
| Unconditional LoRA | ostris/ideogram_4_unconditional_lora |
The gallery images were made at these settings with AI Toolkit, at 1024×1024. The training images also included portrait and landscape shapes, but other resolutions have not been tested yet.
Usage
This runs the default checkpoint through Hugging Face Inference Providers, on your own quota. Set HF_TOKEN to your Hugging Face token, then run:
import os
from huggingface_hub import InferenceClient
client = InferenceClient(provider="fal-ai", api_key=os.environ["HF_TOKEN"])
prompt = '{"high_level_description":"hmrsh painting: a wide view of a woman in a dark dress seen from behind in a sparse room","style_description":{"aesthetics":"quiet, austere, contemplative, still","lighting":"soft diffused light","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#A39A88","#2A2622","#D8D2C4","#6F685A"]},"compositional_deconstruction":{"background":"sparse room with pale walls and a grey wooden floor","elements":[{"type":"obj","bbox":[250,80,950,380],"desc":"woman in a dark dress seen from behind, hair pinned up"}]}}'
image = client.text_to_image(
prompt,
model="jejunepixels/Hammershoi-Ideogram4-LoRA",
width=1024,
height=1024,
seed=42,
extra_body={"expansion_model": "None"}, # stops fal rewriting the JSON prompt
)
image.save("hammershoi.jpg")
fal sets its own step count and guidance, so results won't exactly match the gallery. This route and AI Toolkit are the only ways this LoRA has been tested; loading it with Diffusers has not been tested.
Prompt tips
Ideogram 4 takes a JSON prompt. Keep the keys in this order and write it on one line:
{"high_level_description":"hmrsh painting: a wide view of a woman in a dark dress seen from behind in a sparse room","style_description":{"aesthetics":"quiet, austere, contemplative, still","lighting":"soft diffused light","medium":"painting","art_style":"oil on canvas, danish symbolism, muted palette","color_palette":["#A39A88","#2A2622","#D8D2C4","#6F685A"]},"compositional_deconstruction":{"background":"sparse room with pale walls and a grey wooden floor","elements":[{"type":"obj","bbox":[250,80,950,380],"desc":"woman in a dark dress seen from behind, hair pinned up"}]}}
Framing — use in high_level_description:
wide— full scene with environmentmedium— figure from waist upclose up— head/hands/objectsextreme close up— fine detail cropsfull figure— head to toe
Lighting — choose one for lighting:
overcast daylight— flat, grey outdoor lightsoft diffused light— gentle indoor lightbright diffused light— bright but shadowlessdark chiaroscuro— dramatic shadow and contrastcandlelight— warm, intimate flame lightbacklight from window— silhouette against window
Bounding boxes — optional. bbox is [top, left, bottom, right], each from 0 to 1000. Leave bbox out of an element to let the model place it. In testing, objects generally followed their boxes, though not always exactly.
The same prompt and seed at five training checkpoints, with the bounding boxes drawn on top to show where each object was asked to go. This release is the step 2000 checkpoint.
Always use for art_style:
oil on canvas, danish symbolism, muted palette
Checkpoints
This repo has three checkpoints from the same training run. Later checkpoints have trained longer on the paintings, and each one has its own character, so try more than one. The recommended settings and prompt format are the same for all three.
| File | Step | What we saw in testing |
|---|---|---|
Hammershoi_ideogram4.safetensors (default) |
2000 | The most even balance. Keeps soft fabric such as draped tablecloths. Figures sometimes turn away from the pose you asked for. |
Hammershoi_ideogram4_step2250.safetensors |
2250 | Crisper architecture and stronger contrast between light and shadow. Tablecloths start to lose their drape, and figures can still turn away from the pose you asked for. |
Hammershoi_ideogram4_step2500.safetensors |
2500 | Darker, denser tones, softer faces, and larger figures that fill their boxes. Tablecloths often become hard tops, and it can add details you didn't ask for, such as a distant figure in a doorway. |
The same prompt and seed in each row. These notes come from a small test at 1024×1024 with two seeds per prompt, and the differences are small and vary from image to image.
Hosted inference through Inference Providers uses the default file. To use step 2250 or 2500, download the file and load it in your own workflow.
Limitations and biases
- Testing was small. Checkpoints were compared on 22 prompts with two seeds each, at 1024×1024 only. Other image shapes are untested.
- Boxes and poses are approximate. Objects generally follow their boxes, but not always exactly, and figures can turn away from the pose you asked for.
- Narrow subject range. It was trained on one painter's work: sparse Danish interiors, landscapes and portraits. People in these paintings are mostly European, often women in dark period dress, so expect results to lean that way unless your prompt says otherwise.
- Captions and boxes are not perfect. They were drafted by Claude from the images and earlier captions, then reviewed and corrected by hand. The boxes are close estimates, not pixel-exact.
- Not for commercial use. See the license below.
Training details
| Base model | ideogram-ai/ideogram-4-fp8 |
| Training dataset | jejunepixels/Hammershoi-Ideogram4-Dataset: 288 images with JSON captions and bounding boxes |
| Sources | SMK Open · Wikimedia Commons |
| Training tool | AI Toolkit |
| Training hardware | RTX 4090, using the low_vram setting and qfloat8 quantization |
| Generation speed | About 75 seconds per 1024×1024 image at 30 steps on the same GPU |
| Trigger word | hmrsh |
| LoRA rank | 16 |
| Steps | 2000 (default), 2250 and 2500, from a 2500-step run |
| Learning rate | 0.0001 |
| weight_decay | 0.00001 |
| Sampler | Flowmatch |
| Training quantization | qfloat8 |
| dtype | bf16 |
License
Non-commercial use only. These weights are a fine-tuned derivative of Ideogram 4, modified by jejunepixels, and are released under the Ideogram Non-Commercial Model Agreement. The same terms and use restrictions apply to anyone who uses or redistributes them. A copy of the agreement and the required NOTICE file are included in this repo.
Ideogram 4 is provided under and subject to the Ideogram Non-Commercial Model Agreement available at https://github.com/ideogram-oss/ideogram-4/model_licenses/LICENSE-IDEOGRAM-4-NON-COMMERCIAL. All rights reserved. Copyright © Ideogram, Inc.
This is a community project. It is not an official Ideogram product and has not been endorsed, approved or validated by Ideogram. Images made with this LoRA are AI-generated.
Training images were collected from SMK Open and Wikimedia Commons. The Wikimedia files trace back to museums, auction houses and other sources, and the dataset lists the source of each image. Vilhelm Hammershøi died in 1916 and his paintings are in the public domain; check the source of a reproduction before reusing it.
Download model
Download the weights in the Files & versions tab.
- Downloads last month
- 53
Model tree for jejunepixels/Hammershoi-Ideogram4-LoRA
Base model
ideogram-ai/ideogram-4-fp8