Prompting
CFG scale
CFG scale (classifier-free guidance) sets how strictly a diffusion model follows the prompt. Low values give looser, more varied images; high values follow the words literally at the cost of naturalness.
Around the default the model balances adherence and quality. Push it up for literal compositions (product shots, specific poses); pull it down for painterly or exploratory renders. Very high values over-saturate and stiffen.
Sur eroq
Exposed as `cfg_scale` on the image engines eroq runs itself: 0–3 on Krea2, Eroq Uncensored and Eroq One, 0–10 on Anima, in the studio and the API.
Confidentialité
Privé et sans restrictions
- →Aucun filtre sur les moteurs d’eroq : une politique écrite à la place.
- →Ils ne gardent rien, et rien ne s’entraîne sur ton travail.
- →Le mode privé ne garde aucun fichier.
Questions
What CFG value should I use?
Start at the default; raise it when the render ignores details, lower it when results look rigid.
Which eroq models expose CFG scale?
The image engines eroq runs itself: Krea2, Eroq Uncensored and Eroq One take cfg_scale from 0 to 3, Anima from 0 to 10, in the studio and over the API. The third-party image engines (GPT Image 2.5, Nano Banana, Seedream, Flux, Grok Imagine and Flash Image) and the video engines do not expose it; adherence there comes from the wording and the rack.
Is a higher CFG always better?
No. Past the middle of the range colors over-saturate and poses stiffen — high CFG buys literalness, not quality. Raise it when details from the prompt are missing, not as a habit.
Autres termes de la catégorie prompting
