Rewrite model card in plain language

README.md

@@ -16,9 +16,9 @@ tags:
 
 # SolPix
 
-SolPix generates images from text with a flow transformer of roughly 49M parameters. The transformer works in the latent space of the SANA 1.1 DC-AE, and Flan-T5 Base encodes the prompt. The included helper produces 512x512 images.
+SolPix turns a text prompt into a 512x512 image. Its roughly 49M-parameter flow transformer predicts image latents; Flan-T5 Base encodes the prompt, and the SANA 1.1 DC-AE decodes the result. Both external models stay frozen.
 
-## Generator and encoders
+## Components
 
 | Setting | Value |
 |---|---|
@@ -38,18 +38,18 @@ SolPix generates images from text with a flow transformer of roughly 49M paramet
 | Saved optimizer step | 210,000 |
 | Configured training schedule | 5,000,000 steps |
 
-The 49M count covers the generator. It excludes the frozen text encoder and autoencoder, which are separate dependencies. `SolPixTransformer2D` predicts latent velocity, and `AutoencoderDCSol` loads the matching Diffusers `AutoencoderDC` to decode the sampled latents.
+The parameter count covers the generator only. Downloading the frozen text encoder and autoencoder adds separate dependencies. `SolPixTransformer2D` predicts latent velocity. `AutoencoderDCSol` loads the corresponding Diffusers `AutoencoderDC` for decoding.
 
 ## Generate an image
 
-Download this repository and install `requirements.txt`, then run the helper with a prompt:
+Download the repository, install its requirements, and give `generate.py` a prompt and output path:
 
 ```bash
 python -m pip install -r requirements.txt
 python generate.py --prompt "A glass greenhouse in a quiet garden after rain" --output solpix.png
 ```
 
-You can also set the local checkpoint, seed, and sampling settings:
+To choose a local checkpoint, seed, or sampling settings:
 
 ```bash
 python generate.py \
@@ -59,29 +59,27 @@ python generate.py \
   --output ./solpix.png
 ```
 
-The helper downloads Flan-T5 Base and the pinned SANA DC-AE revision. It samples with Euler integration and classifier-free guidance, using CUDA if available. CPU inference works but is slow.
+On its first run, the helper downloads Flan-T5 Base and the pinned SANA DC-AE revision. It uses Euler integration with classifier-free guidance, running on CUDA when available. CPU inference is supported but slow.
 
-The flow path is `x_t = (1 - t) x_clean + t noise`; sampling integrates from `t=1` to `t=0`. `config.json` records the encoder and decoder identifiers and their revisions.
+The flow path is `x_t = (1 - t) x_clean + t noise`. Sampling runs from `t=1` down to `t=0`; encoder and decoder identifiers and revision pins are in `config.json`.
 
-## Training data and run
+## Data and checkpoint history
 
-The data split comes from [MONET v1.2.0](https://huggingface.co/datasets/jasperai/monet), curated with seed `20260924`. It has 174,603 training examples and a validation holdout of 9,300 examples. Training used pre-encoded SANA F32C32 image latents and Flan-T5 Base caption states.
+We used [MONET v1.2.0](https://huggingface.co/datasets/jasperai/monet) with curation seed `20260924`. The split contains 174,603 training examples and a 9,300-example validation holdout. SANA F32C32 image latents and Flan-T5 Base caption states were encoded before training.
 
-The sources include CC12M and CommonCatalog-CC-BY, COYO, Diffusion-Aesthetic-4K, and LAION. Synthetic captions come from Flux Klein, Flux Schnell, and Z-Image. Curation filters cover resolution and aesthetics, NSFW content, watermarks, and near duplicates.
+MONET draws from CC12M, CommonCatalog-CC-BY, COYO, Diffusion-Aesthetic-4K, and LAION. Flux Klein, Flux Schnell, and Z-Image supply synthetic captions. Curation checks resolution, aesthetics, NSFW content, watermarks, and near duplicates.
 
-Upstream records carry CC BY 4.0, Apache 2.0, Google permissive, and MIT license labels. Those labels describe the source records; they don't grant a new license for the contents. This repository doesn't redistribute the images or dataset shards.
+The source records include CC BY 4.0, Apache 2.0, Google permissive, and MIT license labels. A label on a record doesn't grant a new license to its contents. Images and dataset shards aren't redistributed in this repository.
 
-The Windows v1.0 continuation ran in BF16 on one RTX 3080 Ti, with batch size 4 and gradient accumulation 16. The released checkpoint is step 210,000. The documented 1.1 continuation keeps the same split and targets step 300,000.
+The Windows v1.0 continuation used BF16 on one RTX 3080 Ti, with batch size 4 and gradient accumulation 16. We released optimizer step 210,000. The documented 1.1 continuation retains that split and targets step 300,000.
 
-## Results and limits
+## Reading the samples
 
-The released checkpoint has no formal image-quality or prompt-following benchmark. The gallery lets you inspect generated outputs, but it doesn't provide a held-out quality estimate. Expect composition errors and artifacts, with weak rendering of text or fine detail.
+We haven't run a formal image-quality or prompt-following benchmark on this checkpoint. The gallery shows generated examples, without supplying a held-out quality estimate. Composition errors, artifacts, and weak text or fine-detail rendering remain limitations.
 
-SolPix has no built-in safety classifier. Filtering the training data doesn't remove all source biases or unwanted associations. Flan-T5 and SANA DC-AE also have their own licenses and usage terms.
+There is no built-in safety classifier. Dataset filtering doesn't remove every bias or unwanted association. Flan-T5 and SANA DC-AE have separate licenses and usage terms.
 
-## Generated samples
-
-These 15 images come from the released SolPix checkpoint. Each uses 512×512 resolution, 32 Euler steps, and guidance scale 3.5 with the pinned SANA DC-AE decoder. `samples/` contains the image files and records their prompts, seeds, and SHA-256 values.
+All 15 samples below use the released checkpoint at 512x512, with 32 Euler steps, guidance scale 3.5, and the pinned SANA DC-AE decoder. Their files, prompts, seeds, and SHA-256 values are recorded in `samples/`.
 
 ### Sample 01
 
@@ -188,13 +186,13 @@ Seed: 260939
 Prompt: A small observatory beneath a clear star-filled sky, distant mountains, night landscape photograph.  
 Seed: 260940
 
-## Repository files
+## Files
 
-- `step_00210000.pt` contains EMA and raw weights, optimizer state, configuration, and training arguments.
-- `solpix/` has the transformer and decoder adapter, along with configuration, data, and training components.
-- `generate.py` turns a prompt into an image. `train.py` and `sample_latents.py` are the training and latent-sampling entry points.
-- `samples/` holds the 15 generated PNGs and their metadata. `config.json` records the architecture and external-model manifest.
+- `step_00210000.pt`: EMA and raw weights, optimizer state, configuration, and training arguments.
+- `solpix/`: the transformer, decoder adapter, configuration, data, and training components.
+- `generate.py`: prompt-to-image generation. `train.py` starts training; `sample_latents.py` samples latents.
+- `samples/`: the 15 PNGs and their metadata. `config.json` records the architecture and external-model manifest.
 
 ## License
 
-[Apache 2.0](LICENSE) covers the repository code and checkpoint weights, as well as the configuration, card, and supplied banner. [NOTICE](NOTICE) contains the attribution. The upstream datasets, Flan-T5, and SANA DC-AE keep their own licenses.
+The code, checkpoint weights, configuration, model card, and supplied banner use [Apache 2.0](LICENSE). Attribution is in [NOTICE](NOTICE). Dataset, Flan-T5, and SANA DC-AE licenses apply separately.