SolPix

SolPix is a roughly 49M-parameter text-to-image flow transformer. It generates latents for the SANA 1.1 DC-AE, with text conditioning from Flan-T5 Base.

Model

Setting	Value
Generator	Approximately 49M parameters
Architecture	15-block U-shaped joint text/image transformer
Hidden width	512
Attention	8 heads, head dimension 64
FFN	SwiGLU, width 1,152
Long skip connections	7
Conditioning	Shared adaptive layer normalization
Local mixing	Depthwise 3x3 convolution
Objective	Rectified flow matching with logit-normal time sampling
Image latents	32 channels, 32x spatial compression
512x512 latent grid	16x16
Text encoder	Frozen google/flan-t5-base, 768 features, up to 96 tokens
Decoder	Frozen SANA 1.1 DC-AE F32C32
Saved optimizer step	210,000
Configured training schedule	5,000,000 steps

The text encoder and autoencoder are external dependencies and are excluded from the generator's parameter count. SolPixTransformer2D predicts latent velocity; AutoencoderDCSol loads the matching Diffusers AutoencoderDC to decode sampled latents.

Use

Download this repository and install its requirements:

python -m pip install -r requirements.txt
python generate.py --prompt "A glass greenhouse in a quiet garden after rain" --output solpix.png

To choose a local checkpoint, seed, and sampling settings:

python generate.py \
  --checkpoint ./step_00210000.pt \
  --prompt "A small red sailboat on a misty lake at sunrise" \
  --seed 1234 --steps 40 --guidance-scale 3.5 \
  --output ./solpix.png

The helper downloads Flan-T5 Base and the pinned SANA DC-AE revision, then samples with Euler integration and classifier-free guidance. It produces a 512x512 image and uses CUDA when available. CPU inference is supported but slow.

Sampling follows x_t = (1 - t) x_clean + t noise, integrating from t=1 to t=0. Encoder and decoder identifiers and revisions are in config.json.

Training

The split was curated from MONET v1.2.0 with seed 20260924: 174,603 training examples and a 9,300-example validation holdout. Training used pre-encoded SANA F32C32 latents and Flan-T5 Base caption states.

Sources included CC12M, CommonCatalog-CC-BY, COYO, Diffusion-Aesthetic-4K, LAION, and synthetic captions from Flux Klein, Flux Schnell, and Z-Image. Curation applied resolution, aesthetic, NSFW, watermark, and near-duplicate filters. Source records carried CC BY 4.0, Apache 2.0, Google permissive, and MIT license labels. Those labels describe upstream records, not a new license for their contents. Images and dataset shards are not redistributed here.

The Windows v1.0 continuation used BF16 on one RTX 3080 Ti, with batch size 4 and gradient accumulation 16. The published checkpoint is step 210,000. The documented 1.1 continuation uses the same split and targets step 300,000.

Evaluation and limits

No formal image-quality or prompt-following benchmark accompanies this checkpoint. The gallery shows generated outputs, not a held-out quality estimate. Expect composition errors, artifacts, and weak rendering of text or fine detail.

The model has no built-in safety classifier. Dataset filtering does not remove all source biases or unwanted associations. Flan-T5 and SANA DC-AE have their own licenses and usage terms.

Samples

These are 15 individual images generated by the released SolPix checkpoint, not external examples. Each was sampled at 512×512 with 32 Euler steps and guidance scale 3.5 using the pinned SANA DC-AE decoder. The image files, prompts, seeds, and SHA-256 values are in samples/.

Sample 01

Prompt: Three Black men sharing french fries at a neighborhood diner, candid documentary photography.
Seed: 260926

Sample 02

Prompt: A red fox standing in fresh snow beneath pine trees at winter dawn, wildlife photography.
Seed: 260927

Sample 03

Prompt: A glass greenhouse filled with ferns after rain, soft natural light, botanical photograph.
Seed: 260928

Sample 04

Prompt: A handmade cobalt blue teapot on a pale stone table, clean studio product photograph.
Seed: 260929

Sample 05

Prompt: A white sailboat crossing a calm blue bay at golden hour, fine art landscape photograph.
Seed: 260930

Sample 06

Prompt: An orange cat curled on a wooden chair in a sunlit bookshop, cozy editorial photograph.
Seed: 260931

Sample 07

Prompt: A small street cafe reflected in wet pavement at night, warm window light, city photograph.
Seed: 260932

Sample 08

Prompt: A wooden lighthouse on a rocky coast under a cloudy sky, atmospheric landscape photograph.
Seed: 260933

Sample 09

Prompt: A bowl of ripe peaches on a kitchen counter, morning light, natural still life photograph.
Seed: 260934

Sample 10

Prompt: A snow-covered cabin among tall pine trees at blue hour, quiet winter landscape photograph.
Seed: 260935

Sample 11

Prompt: A baker placing fresh bread on a cooling rack in a bright kitchen, documentary photograph.
Seed: 260936

Sample 12

Prompt: A goldfinch perched on a thin branch among spring blossoms, close-up wildlife photograph.
Seed: 260937

Sample 13

Prompt: A red bicycle leaning against a brick wall on a leafy neighborhood street, lifestyle photograph.
Seed: 260938

Sample 14

Prompt: A lemon cake with a slice cut out on a ceramic plate, bright tabletop food photograph.
Seed: 260939

Sample 15

Prompt: A small observatory beneath a clear star-filled sky, distant mountains, night landscape photograph.
Seed: 260940

Files

• step_00210000.pt: EMA and raw weights, optimizer state, configuration, and training arguments.
• solpix/: transformer, decoder adapter, configuration, data, and training components.
• generate.py: prompt-to-image helper.
• train.py and sample_latents.py: training and latent-sampling entry points.
• samples/: 15 generated PNGs and their metadata.
• config.json: architecture and external-model manifest.

License

Repository code, checkpoint weights, configuration, card, and supplied banner are licensed under Apache 2.0. See NOTICE for attribution. This license does not relicense the upstream datasets, Flan-T5, or SANA DC-AE.