Models & Parameters

CFG Scale: How Hard to Push the Prompt

Also called what is cfg scale, guidance scale, classifier free guidance, cfg, prompt strength

CFG scale is the strength of classifier free guidance, the setting that decides how far a diffusion model is pushed toward your prompt and away from what it would have drawn unprompted. Low values give loose, soft results, high values give literal, contrasty ones, and every model family has its own working range.

What the number actually controls

At every denoising step a diffusion model quietly makes two predictions: one conditioned on your prompt, one conditioned on nothing at all. Classifier free guidance takes the difference between them and amplifies it. The scale is the amplifier. At 1 you get the conditioned prediction untouched. At 7 you get the prompt direction exaggerated sevenfold, so the output leans harder into whatever makes it match your words.

That framing explains everything else on this page. The setting does not add information. It exaggerates a direction the model already computed, and exaggeration has a ceiling before it turns into distortion.

Ranges that work

Model familyTypicalNotes
SD 1.56 to 9Tolerates high values, needs a negative prompt
SDXL5 to 87 is a safe default
Flux dev style2.5 to 4Distilled guidance, different meaning
Turbo / lightning1 to 2Effectively fixed, do not raise
Video models4 to 7High values amplify motion artefacts

Treat these as starting points, not settings. The only reliable method is to lock the seed, change guidance alone, and compare a small ladder of values such as 3, 5, 7, and 9 on the same prompt.

Reading a bad result

Too low looks under committed: washed out colour, soft edges, a composition that only half matches the prompt, backgrounds that drift into abstraction. People usually diagnose this as a weak prompt and start adding words, which makes it worse.

Too high looks fried: clipped saturation, blown highlights, dark outlines around subjects, plastic skin, and mangled hands. In video it shows up as jitter and popping, because guidance is being exaggerated independently on frames that need to agree with each other.

Both extremes are easier to spot side by side than alone, which is the argument for running the ladder once per model and remembering the answer.

When it cannot help

Three situations where turning the dial is wasted effort. First, missing knowledge: if the checkpoint has never learned a concept, no amount of amplification will produce it. Second, dropped clauses: a prompt with six competing instructions loses some of them to attention dilution, and guidance amplifies the surviving direction rather than restoring the lost ones. Third, distilled models, where the value is fixed by training and the field is decoration.

The practical habit worth building: treat CFG scale as a per model constant you find once, then leave alone. Prompt content, seed, and steps are the things you actually vary run to run. Almost nobody needs to touch guidance twice on the same model, and treating it as a creative control is how sessions turn into forty generations of the same burned image.

The prompt for this

A starting point that reliably produces the effect. Adjust the subject and setting; keep the technical clauses.

A brass diving helmet on a wooden workbench, single window light, dust in the air, muted colours, 35mm film

Try CFG Scale yourself

Open the generator with a starting point already filled in.

Frequently asked questions

What is a good CFG scale?
There is no universal number, only a number per model family. Around 6 to 8 for SD 1.5 and SDXL, 2.5 to 4 for Flux style distilled guidance, 1 for turbo and lightning checkpoints, and 4 to 7 for most video models. Change one setting at a time and judge on a fixed seed.
Why do my images look oversaturated and burned?
That is the classic overshoot from guidance set too high. Colour clips, highlights blow out, edges get a halo, and faces go waxy. Drop by two and regenerate on the same seed. If you need contrast, get it from lighting words in the prompt rather than from guidance.
Does CFG scale work on Flux or turbo models?
Not the same way. Distilled models have guidance baked in during training, so they expose either a much lower dial with a different meaning or none at all. On a turbo checkpoint anything above about 2 usually degrades the image instead of sharpening the prompt.
What is the relationship between CFG scale and sampling steps?
High guidance takes larger jumps toward the prompt, which needs more steps to stay stable. If you raise guidance and results turn crunchy or noisy, add steps before assuming the model is wrong. Low guidance is more forgiving and converges in fewer steps.
Is guidance scale the same thing as CFG scale?
In most interfaces yes, they are the same dial under two labels. The exception is distilled models, where a field named guidance is a learned embedding value rather than true classifier free guidance, so the familiar range does not transfer.
Does raising it make the model follow more of my prompt?
Up to a point. Beyond it the model follows the prompt more literally but not more completely: dropped clauses stay dropped, because that is an attention problem. If a clause is being ignored, shorten the prompt instead of pushing guidance higher.

Related terms