---
license: apache-2.0
---
<h1>RealFlux 1.0b Model Training (Third Attempt)</h1>

<h2>Stage 1: Data Preparation</h2>
    <ul>
        <li>[ ☑ ] Collect images for the dataset</li>
        <li>[ ☑ ] Image captioning</li>
    </ul>

<h2>Stage 2: Model Training</h2>
    <ul>
        <li>[ ☑ ] Prepare dataset for model input</li>
        <li>[ ☑ ] Model training</li>
    </ul>

<h2>Stage 3: Testing</h2>
    <ul>
        <li>[ ☑ ] Schnell Model testing</li>
        <li>[ ▶ ] Dev Model testing</li>
    </ul>

<h2>Stage 4: Release</h2>
    <ul>
        <li>[ ☑ / ▶ ] Release Schnell Compact Version and Transformer</li>
        <li>[ ☐ / ☐ ] Release Dev Compact Version and Transformer</li>
        <li>[ ☐ ] Release GGUF Q8 version</li>
    </ul>

<h2>Comments</h2>

<p>The first version of the model turned out to be unsuccessful due to significant anatomical errors. The second version of the model is being trained on a reassembled dataset with better-chosen training parameters.</p>
  
<p>The second version of the model has even fewer anatomical errors and text errors. The model has been sent for refinement with a slightly modified dataset.</p>

<h2>Limitations of RealFlux 1.0b Schnell</h2>
    <ul>
        <li>Cannot generate NSFW content (the ability to generate basic NSFW content will be added in future versions).</li>
        <li>There may be issues with anatomy, as the model is based on Flux Schnell.</li>
        <li>There may be issues with generating images in the anime style (the quality of generations will be improved in future versions).</li>
    </ul>