SpicyAPI

One frame in, fifteen seconds of video with sound out

Hold a cast, a wardrobe, or a room steady across a shot

Getting a watchable clip out of nothing but a prompt

Narration with tone instructions that actually do something

Instruction editing with an optional mask for the pixels you care about

Edit a photo in a style you trained into a LoRA

Your own LoRA, by URL, no download required

The open-weight Qwen you can't be bothered to load locally

Camera angles and relights without training anything

The base refresh that actually likes your LoRAs

One image, one instruction, and a shape you have to choose yourself

The short-prompt Qwen, and why the version numbers lie to you

Merge references and change the words already in the picture

Retype the cover line without redrawing the cover

Pay the extra cent when skin, glass and typography have to hold up

The node you use when the image needs readable text in it

Three inputs, no mask, and the least polite editor in the pack

Put your character's face on someone else's dance

The open model, minus the 25-minute render

Separate LoRA slots for the high-noise and low-noise experts

Attach your own weights to the one Wan you can still download

The uncensored endpoint, and why 'Spicy' isn't Alibaba's word

Uncensored endpoint, your weights on top

Adding seconds to a clip you already have

The cheap baseline, and what you give up for it

When you need the audio, not just the motion

Text in, ten seconds of video with sound out

The cheap tier, and why it has no soundtrack

15 seconds with sound, and a multi-shot switch nobody uses

Cast a subject from a clip, not a photo

Uncensored, with the 15-second runway

Write the shot list, get 15 seconds with sound

Edit up to nine reference images with one instruction

The 2k canvas, for edits with several moving parts

4k canvases and a colour palette you actually control

Uncensored, up to 15 seconds, with a negative prompt

The reasoning pass that plans the shot before drawing it

30 seconds, and a '-1' that lets the model pick the length

Same schema as 3.0, about a third more per second

Hand it a script, a web page and five clips

One frame in, an unfiltered clip out

Prompt-only video with sound, on someone else's GPU

Animate a still at 2K and 4K (which are upscales)

The expensive tier, and when it earns it

Lock a face, a garment and a voice

4K video from a sentence, at $0.234 a second

Keep the same person, product or place

The mid-tier that's usually the smart buy

The cheapest way to keep a subject consistent

Animate a still without the filter conversation

The cheap drafting node you'll end up keeping

Restyle a photo with your own LoRAs

Test your trained LoRA for a cent and a half

You set the pixels, up to 2560 a side

Exact pixel control for about a penny

Restyle a photo in about a second's worth of compute

Your LoRA, hosted, for a cent a picture

A penny a picture, and the cheapest good frame in the graph

The revival act, with a mask that actually works

Borrow a 12B LoRA library without the 12B download

The still you already made, now moving

Same shot, fewer refusals

A whole shot with sound from one prompt

What '-1' does, and why audio is free here

No first frame, just your own material

Uncensored, and audio costs nothing extra

The switch that makes it search the web first

The one that goes to 4K

Seedance 2.0 for about two cents a second

The cheapest way to point at your own footage

The uncensored option that doesn't cost the earth

No image, no GPU, cheap takes

Nine images, three clips, three tracks, one shot

Uncensored, and it goes to 4K

Seven ratios, 4K, and what web search really costs

30-second clips, and why 'adaptive' is your only ratio

30 images of your own material is the whole trick

The priciest node in the Seedance line

Half a minute of footage from one prompt

One prompt, a whole scene, several speakers

Pick the exact pixels before you describe the change

One prompt, sixteen exact sizes, one flat rate

Ten references, fifteen ratios, and a real alpha channel

Turn a flat PNG into sixteen movable layers

Fifteen ratios, three resolutions, one price

Eight exact sizes, one uncensored edit model

A 2K frame from a short prompt, at eight fixed sizes

Thinking mode, ten references, real transparency

Split a finished image into named, movable layers

The thinking pass is free, so leave it on

41 voices, knobs for speed and pitch, no filter

25 text models in one node, writing your prompts for you

The strength dial that decides how much of your photo survives

Restyle a photo with weights you already trained

Your anime LoRAs, rented by the image

The cheapest anime node in the pack, and the most literal

30 voices, acted-out cues like [sighs], and two-speaker dialogue

The cheapest voice in the pack, with one caveat

30 voices, cues that get acted out, half the Pro price

Fourteen reference images, and it can search the web mid-render

4K for twelve cents, and it can Google your prompt first

Stack up to 14 reference images and tell Google what to change

The model that plans the poster before it draws it

Get the lyrics back out of a song you generated

Bring your own weights for three cents, without the fp8 headache

Two and a half cents for a photoreal portrait, no GPU involved

The cheapest way to make a still move

1080p motion at twenty-two cents a second, and you'd better mean it

$2.16 for ten seconds, silent, 1080p

Sound on every clip, whether you want it or not

1080p with sound for twenty-eight cents a clip

Put your character's face on someone else's dance

1080p, sound optional, and the audio toggle doubles the price

Describe the scene, decide whether you want sound

The version with knobs — orientation, shots, original audio

3 to 15 seconds, 4K if you're feeling wealthy, and a real shot list

Write a shot list, get a multi-shot clip back

You pay more for less, and that's the trade

Four widgets, sound always on, no negative prompt

One portrait, one audio file, and the picture talks

Animate one still, and keep an eye on the meter

Seven photos, one shot, characters that stay put

No image, no seed, and no free retries

Change one thing in a clip and leave the rest frozen

The finishing tier, complete with reference images and anamorphic frames

The volume tier, and the only Krea with a format switch

A penny and a half per idea, ten style references included

All the LTX, none of the 22GB text encoder

Your own weights, no GPU, no VAE archaeology

Stack three of your own weights on a hosted LTX

20 seconds with a soundtrack for under thirty cents

The release that fixed frozen clips, without the filter

Thirteen built-in LoRAs, and the one input that will confuse you

The architectural jump, up to 4K, with an ending you choose

Long single takes at 4K, and a soundtrack you don't have to ask for

It never loads a LoRA — it just points the server at a file

Put the same face in a new scene, from one prompt

Four megapixels, your weights, four cents

The audio-native model, at two and a half cents a second

Your own weights on an audio-native model

Nine Images, Three Clips and Your Own LoRAs in One Shot

Pick Your Own Weights Without Leaving ComfyUI

Give It a First Frame, Get the Shot With Sound

Fast H3 Clips, and Mind That 1:1 Default

Nine Images In, One Character Who Actually Stays Put

The Community Fine-Tune, With Your LoRAs Stacked On

Nine Images, Three Clips, Three LoRAs

An Unfiltered H3 Build for the Frames Others Refuse

2K Clips With Real Sound, From About 2.5 Cents a Second

A Finished Song for Eighteen Cents, Vocals Included

472 Voices, an Emotion Dial, and (sighs) Actually Performed

The Cheap One. For Most Scripts, Take It.

Booru Tags or Plain English, Twelve Bucks a Thousand Images

Sixteen Sources, One Mask, and an Alpha Channel You Can Actually Use

The One You Use When the Picture Needs Legible Words

Same Inputs as Flare, Slower, Better, Worth the Wait

4K, Legible Type, and a Local Alternative Muscling In

Ten Sources, and an `auto` Aspect Ratio That Just Keeps Your Frame

Medium Quality, Legible Type, and Fifteen Aspect Ratios

Booru Tags, Score Tags, and Your Pony LoRAs, Rented by the Image

Pony's prompt dialect without the 6 GB download

The open-weights TTS everyone clones voices with, no install

You want the alpha mask, not the flat image

The analysis pass that doesn't touch your video

One JSON list and the person is someone else

$0.16 a shot when the cheap swap isn't good enough

The thirteen-cent swap, and what it won't do

Skin tone is 90% of why swaps look fake

Keeping one face on the same head for 30 seconds

The last stage your silent render has been missing

When the hair has to come with the face

Find every head in the clip before you change one

The swap call, and the two values that must match

Outpaint to a new aspect ratio without painting a mask

8K costs the same as 2K, so think about which you're asking for

Redub the mouth, keep the performance

The weird one, and why it can't be seeded

Name the thing, and it's gone

One portrait, one voice clip, one video

The cheap way to make a 480p render look finished

Two Finished Songs for Seven Cents

Hand It Your Lyrics or Let It Write Them

The Suno Node You Use When the Safe One Sounds Boring

Two-Cent Seconds, and the Prompt Is Optional

Cheap, Silent, and Weirdly Unfiltered

Sound Included, at 1080p and Seven Cents a Second

Cheaper Per Second, and Permanently 540p

The Unfiltered Vidu Node, and What It Doesn't Have

The Video Node You'd Actually Draft With

Two Hours of Audio Transcribed for About a Quarter

71 Voices, and One Setting That Will Ruin Your Audio