parameters

Aspect ratio of the output. Accepts `w:h` or one of the named aliases.

default

model-specific (usually 1:1)

values

1:12:33:24:55:49:1616:921:9squareportraitlandscapewideultrawidevertical

supported by

all image modelsmost video models

examples

cinematic portrait --ar 16:9phone wallpaper --ar 9:16poster art --ar 2:3

How many variations to return. `--1` through `--4`. Default is model-specific (usually 4 for txt2img, 1 elsewhere).

values

--1--2--3--4

supported by

all image modelsall video models

examples

a cyberpunk dragon --2quick sketch --1

Freeze the RNG. Same seed + same prompt + same model = same image.

range

0 – 4294967295

supported by

all image modelsall video models

examples

forest at dusk --seed 42portrait --seed 99999

Anime / illustration bias. Optionally consumes a version integer (`--niji 6`).

default

5 when no version is given

values

--niji--niji 4--niji 5--niji 6

supported by

anime-tuned models (auto-selected when --niji is present)

examples

catgirl mage --niji 6ramen shop street scene --niji

Suppress the platform’s cinematic default styling. Closer to the underlying model output.

supported by

models that honor the raw style hint

examples

pass photo --raw

Free-text style hint. Provider-specific; also accepts a small enum on Ideogram-family models.

values

rawcuteexpressiveoriginalscenic…or free text (≤50 chars)

supported by

Ideogram v3Recraft

examples

a fox --style expressivebrand mark --style "clean bauhaus"

Capability flag: filter the candidate model pool to ones that produce native alpha / transparent backgrounds. Stacks with `--flagship`, `--sora`, etc.

supported by

Ideogram v3 TransparentRecraft SVG family • ops: txt2img

compat

Stackable with model triggers.

Empty intersection (e.g. `--transparent` + a model that never supports alpha) → explicit error.

examples

app icon on transparent background --transparent

Style reference. Pass one or more image URLs; each may carry a per-URL weight via `::N`.

range

weight per URL: 0 – 1000

supported by

FofrStyleTransfer + style-reference-capable models • ops: txt2img, img2img

compat

Silently ignored by models that don’t support style-reference (no error today).

Do NOT combine with `--vary` (mutually exclusive; not currently validated).

examples

a knight --sref https://cdn.example.com/style.jpgancient temple --sref https://a.jpg::500 https://b.jpg::200

Global style-reference weight. Used when style-reference-capable model is picked.

range

0 – 1000

supported by

style-reference-capable models

examples

portrait --sref https://x.jpg --sw 400

Face reference URL for face-restoration operations.

supported by

face-restore-capable models • ops: img2img

examples

restore face --restoreface --face https://cdn.example.com/me.jpg

Free-text region-of-focus hint for editing operations.

examples

edit the eyes only --focus "left eye"

Force PNG output.

supported by

most image models

examples

icon --png

Force JPEG output.

supported by

most image models

examples

photo --jpg

Force WebP output (used to preserve alpha end-to-end with `--transparent`).

supported by

most image models

compat

If the provider can’t emit WebP, we fall back to PNG.

Force vector SVG output.

supported by

Ideogram v3 (SVG-capable)Recraft SVG family

examples

logo mark --svg

High-resolution image tier (~2048px on long side). Costs more credits.

supported by

Nano-Banana ProFlux 2 ProSeedream 4 • ops: txt2img, img2img

examples

product hero shot --ar 16:9 --2k

Ultra-high-resolution image tier (~4096px on long side). Costs more credits.

supported by

Nano-Banana ProFlux 2 ProSeedream 4 • ops: txt2img, img2img

examples

cinematic still --ar 16:9 --4k

Run background removal on an input image.

supported by

bg-removal models • ops: img2img

Generate directly with a transparent / no-background result.

supported by

transparent-capable models

Restoration pass on a degraded input image.

supported by

restoration models • ops: img2img

Face-only restoration pass. Pair with `--face <url>` for identity guidance.

supported by

face-restore models • ops: img2img

compat

Usually combined with `--face <url>`.

Combine multiple input images into a single scene.

supported by

multi-input img2img models • ops: img2img

Upscale an input image (`upscale_2x` operation).

supported by

Clarity Upscaler et al. • ops: upscale

Refinement pass on a generated image.

supported by

refinement models • ops: img2img

Create a variation of a generated image (`vary_strong` operation).

supported by

variation-capable models • ops: img2img

compat

Mutually exclusive with `--sref` (no engine-side validation today).

Subtle variation of a generated image.

supported by

variation-capable models • ops: img2img

Strong variation of a generated image.

supported by

variation-capable models • ops: img2img

Extend the canvas to the left (outpainting).

supported by

outpaint-capable models • ops: outpaint

compat

Cannot combine with `--ar` in the same call.

Extend the canvas to the right (outpainting).

supported by

outpaint-capable models • ops: outpaint

compat

Cannot combine with `--ar` in the same call.

Extend the canvas upward (outpainting).

supported by

outpaint-capable models • ops: outpaint

compat

Cannot combine with `--ar` in the same call.

Extend the canvas downward (outpainting).

supported by

outpaint-capable models • ops: outpaint

compat

Cannot combine with `--ar` in the same call.

Zoom out (1.5×) — original stays; new content fills the border.

supported by

outpaint-capable models • ops: outpaint

Zoom out at 2× — same as `--zoomout` but wider border.

supported by

outpaint-capable models • ops: outpaint

Zoom-out variant that expands the composition rather than just adding border.

supported by

outpaint-capable models • ops: outpaint

2× expand-out variant.

supported by

outpaint-capable models • ops: outpaint

Dynamic extend amount for outpainting. Two syntaxes: `--0.5` (fraction of edge) or `--512px` (fixed pixel border).

range

fraction: 0.01 – 1.0 • pixels: 1 – 2048

supported by

outpaint-capable models • ops: outpaint

examples

--pan_right --0.5--pan_down --512px

Trigger video generation. If input images are present it becomes img2vid; otherwise txt2vid.

supported by

Kling v2.1Sora / Sora 2 Pro / Sora LiteSeedance 2 (Fast/Pro) • ops: txt2vid, img2vid, ref2vid

examples

clouds rolling over mountains --mp4 --5sec --720p

Video resolution flags. Also accepted with `24fps` / `30fps` suffix.

values

--480p--540p--720p--1024p--1080p--480p24 / --480p24fps--540p24 / --540p24fps--720p24 / --720p24fps--1024p30 / --1024p30fps--1080p24 / --1080p24fps

supported by

Kling v2.1 (720p, 1080p)Sora / Sora 2 ProSeedance 2 (Fast/Pro)

examples

neon city --mp4 --1080p

Frame rate. Not all providers accept every rate — 24 and 30 are the safest.

values

--24fps--30fps--50fps--60fps

supported by

most video models

examples

drone shot --mp4 --24fps

Clip length. Accepts either `--5s` or `--5sec` — both work.

values

--4s | --4sec--5s | --5sec--6s | --6sec--8s | --8sec--10s | --10sec--12s | --12sec

supported by

Kling v2.1 (5s, 10s)Sora models (variable)Seedance 2 (Fast/Pro)

examples

drone shot over reef --mp4 --10sec --1080p

Force the flagship (highest-quality) model of the current operation category. Universal opt-in.

supported by

auto-picks per operation

examples

brand ad --ar 16:9 --flagshipcinematic clip --mp4 --flagship

Force OpenAI Sora / Sora 2 Pro (video).

supported by

Sora 2 Pro • ops: txt2vid, img2vid

compat

Only one model trigger per prompt.

examples

glass shattering slow-mo --mp4 --sora2pro --1080p

Force OpenAI Sora Lite (video). Cheaper than Sora Pro.

supported by

Sora Lite • ops: txt2vid, img2vid

Force Google Nano-Banana Pro. Single trigger claims both txt2img and img2img variants — router picks by op.

supported by

Nano-Banana Pro • ops: txt2img, img2img

examples

product mockup --nanobananapro --2k

note

Was `--nanobanapro` (typo) for ~1 day before 2026-06-01 rename.

Force Google Nano-Banana 2. Trigger claims txt2img + img2img.

supported by

Nano-Banana 2 • ops: txt2img, img2img

Force OpenAI GPT-image-2 (medium quality). Trigger claims both txt2img + img2img variants.

supported by

GPT-image-2 Med • ops: txt2img, img2img

Force GPT-image-2 High quality. Costs more; slower.

supported by

GPT-image-2 High • ops: txt2img, img2img

Force ByteDance Seedance 2 (Fast tier, video). Claims txt2vid + img2vid + ref2vid.

supported by

Seedance 2 Fast • ops: txt2vid, img2vid, ref2vid

Force ByteDance Seedance 2 Pro (video). Claims txt2vid + img2vid + ref2vid.

supported by

Seedance 2 Pro • ops: txt2vid, img2vid, ref2vid

Logged-in-only flag that lets an account access seeded demo generations without spending real credits.