Skip to main content

Settings Cheat Sheet

← Quick reference
Quick lookup of defaults, ranges, and when to change each setting. Full per-model context lives in the Model catalog.

Universal settings (Mago video models)

Defaults and ranges below reflect the current model configs: universal settings from Mago V5, Character settings from Mago V4, Upscaler settings from the Mago upscaler model.
SettingDefaultRangeWhen to change
Output size1280512–1920Lower for testing. Higher for final delivery. 1280 is the sweet spot.
Steps44–20More for sharper detail (costs more). Lower for rougher styles.
InterpolationOffOff/OnOn for cheaper, faster renders. Off for detail-critical work. Exception: Mago Inpaint defaults to On (less flicker in the masked region).
Image sequence exportPNGPNG / EXR 16-bitEXR for professional VFX pipelines.
Prompt influence (CFG)10.5–6Lower if the prompt over-influences. Higher if the model ignores it.
SeedRandom0–4294967295Fix to reproduce a specific result.
Context size15024–300Lower for high movement. Higher for static.
Context overlap100–24Lower for high movement. Higher for slow shots.
Dynamic referenceOnOn/OffOff only for very static shots.
Start frame buffer11 or 5Increase if flicker appears at render start.

Mago Character specific

SettingDefaultWhen to change
Pose strength1Lower for exact tracking. Higher for stylistic freedom.
Face strength1Lower for precise expression replication. Higher for stylized.
Face crop resolution512768/1024 for high-res output or eye issues.
Face crop padding1010–20 if facial features clipped.
Masking threshold0.30.1–0.2 if detection failing. 0.4–0.6 if too much detected.
Masking promptcharacterMulti-subject shots need specifics.
Grow mask1015–25 for spiky hair, horns, flowing clothes.
Relight1Slider 0–1.5. Lower toward 0 to reduce lighting correction; above 1 strengthens it but may affect other aspects.

Mago Style Transfer specific

SettingDefaultWhen to change
Video input strength0.5Below 0.5 for shots over 80 frames. Higher to lock to source.

Creative Upscaler specific

SettingRangeWhen to use
Denoise0.1–0.90.3 light detail · 0.5 moderate · 0.7–0.8 heavy restoration of damaged input.
Tile width / height countHigher for detailRisk of visible tiling at high counts.

Model decision tree

Follow the questions in order.

Video work

Image work (Q5: what are you doing with the image?)

  • Edit a specific element → GPT Image 2 (use the mask image input if region-specific).
  • General style transfer or transformation → Nano Banana Pro.
  • Need 4K output → Nano Banana 2.
  • Character preparation matching source pose → GPT Image 2 or Nano Banana Pro.

Upscale work (Q6: is the input already what you want, just the wrong size?)

  • Yes → Upscaler at ×2 or ×4.
  • No, needs reconstruction or detail → Creative Upscaler. Set denoise by how much reconstruction you want.

Single pass or multi-pass? (Q7: how complex is the transformation?)

  • Simple, single change → single pass with the right model.
  • Multiple changes (character + style + upscale)multi-pass using Edit this Render between stages.
  • Long-form, production work → always multi-pass. The control is worth the extra time.

← Quick reference