We made numerous backend changes to support the full Checkpoint Coverage rollout, and while this is a huge step forward, we recognize that it hasn’t been a smooth transition. Updating to newer versions has caused some systems to behave differently than before, leading to unexpected image quality issues and stability challenges. We know this has been frustrating, and we sincerely appreciate your patience as we work through these challenges. While we’ve already rolled out several fixes and optimizations, we’re not done yet - stabilizing the system and improving generation quality remain our top priorities.
One of the key adjustments we’ve made is a change to the prompt parser - a shift that brings greater accuracy and consistency to generations moving forward.
Prompt Parser Change
In our ongoing effort to improve generation quality, we’ve made a significant backend change: we’ve moved away from Automatic1111-style prompt parsing and adopted ComfyUI’s method.
So what does this mean, and why does it matter?
The Difference Between Automatic1111 and ComfyUI Parsing
Automatic1111’s prompt parser applied a form of mean normalization to weights, effectively averaging their values across the entire prompt. This implementation required a “heavy-handed” approach to weighting, where users had to compensate with exaggerated values to get desired effects.
ComfyUI’s approach treats each token weight individually rather than averaging them. This method is more precise, preserving the original intent of your prompt more faithfully.
Why We Made the Switch
Inconsistent Results: With our updated systems the Automatic1111 Parser was outputting extremely distorted imagery.
More Control: Your weight values will now behave exactly as you specify, allowing for greater precision in prompt crafting.
Future-Proofing: ComfyUI’s parsing is a more modern, actively maintained system (Automatic1111 hasn’t been updated in over 8 months!) that aligns better with evolving generation standards.
Prompt Examples
An example of a prompt which previously may have worked due to Automatic1111's normalization of prompt weights;
highly detailed portrait of a wise old rabbit king (crown:1.4) in a (dark woolen cloak:1.5) and blue glowing magical necklace, (fur:1.6), deep dark forest background, (shallow depth of field:1.4), foliage, ((global illumination)), ((radiant light)), detailed and intricate environment #cinematic, light shafts, dust motes, sparkles, (Extremely Detailed Oil Painting:1.5)
A prompt with refined weighting - a lighter touch is required;
highly detailed portrait of a wise old rabbit king (crown:1.2) in a (dark woolen cloak:1.1) and blue glowing magical necklace, fur, deep dark forest background, (shallow depth of field:1.1), foliage, global illumination, radiant light, detailed and intricate environment cinematic, light shafts, dust motes, sparkles, (Extremely Detailed Oil Painting:1.2)
Additionally, there are a number of good prompting principles which should be followed to get the most out of your generations;
Remixing Old Prompts
We've updated Remixing to make sure older, heavily weighted prompts can be easily re-used without causing unexpected results. Now, when you Remix an image, all weighting and emphasis in the positive and negative prompts will be automatically removed for a cleaner, more predictable generation.
Want the original weights back? A “Restore Original Prompt Weights” link lets you bring them back with a single click;
This update should resolve many of the inconsistencies users have reported. If you experience any unexpected changes in output, we encourage you to tweak your prompt weights accordingly – since the system is now handling them more literally, small weight adjustments will yield better results than before.
If your generations still aren’t turning out right, please let us know. You can submit a Support Ticket via our Ticket Portal, or email [email protected].