Improving Thumbnails
Your first generated thumbnail is a starting point, not the finish line. Subscribr gives you several tools to iterate and refine until you have a thumbnail you love.
Reviewing Generated Thumbnails
After generating a thumbnail, the detail panel shows your result along with:
- The current thumbnail -- The most recently selected variation.
- Thumbnail variations -- All versions generated for this idea, displayed as a gallery. Click any variation to select it as the current thumbnail.
- The concept description -- The creative direction the AI used, which you can edit for future renders.
Take a moment to evaluate what works and what does not. Is the composition strong but the colors off? Is the text placement good but the overall mood wrong? Identifying specific things to change leads to much better improvements.
Providing Feedback for Improvements
When you see something you want to change:
- Click Improve in the detail panel.
- Enter your feedback in the text field. Be specific about what you want changed.
- Optionally attach a reference image to show the AI what you mean.
- Click Generate Improved Thumbnail.
The AI uses your feedback alongside the previous version to generate an improved thumbnail. It builds on the existing result rather than starting from scratch, so your feedback is cumulative.
Writing Effective Feedback
Specific feedback produces better results than vague directions:
| Less effective | More effective |
|---|---|
| "Make it better" | "Make the text larger and add a yellow glow effect" |
| "I don't like the colors" | "Switch to a dark blue background with white text" |
| "More exciting" | "Add more contrast and make the facial expression more surprised" |
| "Fix the layout" | "Move the text to the left side and make the person larger on the right" |
Focus on one or two changes at a time. Asking for too many changes in a single round can dilute the results.
Use Visual Language
Describe what you want to see, not abstract concepts. "Add a bright orange border around the text" is easier for the AI to act on than "make the text stand out more."
How Re-generation Works
Each time you generate an improved thumbnail, the AI:
- Looks at the previous thumbnail as a baseline.
- Reads your feedback to understand what to change.
- Checks any reference images you have attached.
- Generates a new version that applies your requested changes while preserving what worked.
This means improvements are iterative. The third version builds on the second, which built on the first. Your feedback history creates a progressively refined result.
Thumbnail Variations
Every generation creates a new variation that is added to your gallery. This gives you a visual history of all the versions created for this idea.
- Browse variations by scrolling through the gallery in the detail panel.
- Select a variation by clicking it. This sets it as the active thumbnail.
- Download any variation using the download button.
Variations are stored permanently, so you can always go back to an earlier version if a later iteration went in the wrong direction.
AI-Powered Auto-Improve
For hands-free refinement, use the Auto-Improve feature. This lets the AI analyze your current thumbnail, identify what could be better, and generate improved versions automatically.
How Auto-Improve Works
- Open a generated thumbnail in the detail panel.
- Click Auto-Improve.
- Select how many improvement iterations you want (each iteration generates one new thumbnail).
- The AI analyzes the current thumbnail, writes its own feedback, and generates an improved version. It then repeats this process for each iteration.
During auto-improve, you can watch the progress in a timeline that shows each iteration's status and the AI's analysis. Each iteration uses one generation from your quota.
When to Use Auto-Improve
Auto-improve works best when:
- You have a solid starting thumbnail but want to push the quality higher.
- You are not sure what specific feedback to give.
- You want to quickly generate several variations to choose from.
For major direction changes (like switching from a dark mood to a bright one), manual feedback is more effective because you can tell the AI exactly what you want.
Adding Reference Images During Improvement
You can attach reference images at any point during the improvement process:
- Before your first render -- Set the visual direction from the start.
- During improvement rounds -- Add new references to guide specific changes.
Each reference can have a usage label (style, face, product, layout) and an optional custom description explaining how the AI should use it. Up to 5 references can be active at once.
Tips for Getting Better Results
Start with the right mode. If you know the style you want, Clone mode gives you a head start. If you want to explore options, Brainstorm mode offers more variety.
Set up channel defaults. Upload your face photo, logo, and other recurring assets in thumbnail settings. These are automatically available during generation and save you from re-uploading each time.
Iterate in small steps. Change one or two things per round rather than requesting a complete overhaul. Small, targeted changes are more predictable.
Use the concept description. In Brainstorm mode, you can edit the concept description in the detail panel. Changing this text before generating adjusts the creative direction without needing to go through feedback.
Try multiple concepts. If your first concept does not produce good results after a few iterations, switch to a different concept sketch from the same batch. Sometimes a different composition works better for your topic.
Next Steps
- Configure thumbnail settings to set up default assets and style preferences.
- Understand credits and usage to plan your generation budget.