PopNow
PopNow for ComfyUI: every PopNow AI video, image and audio model as a node (Kling, Seedance, Wan, MiniMax, Nano Banana, GPT Image, Flux, Suno). No GPU needed.
Nodes (307)
A still, a sentence, and 15 seconds of clock
Keep the character, lose the first-frame leash
The version with no seed and no excuses
Point it at a script and describe the voice like a person
Your own style LoRA, no mask, four and a half cents
The typography model, now with your style attached
Ten reference photos, a real alpha channel, and a mask if you want one
Ten references and three of your own weights at once
Cheaper than the edit, same three LoRA slots
Six 2K edits per request, and a thinking switch that does nothing until you flip another
Six 2K candidates a pop, and only when you ask for them
2K, transparent backgrounds, and a prompt rewriter that's on by default
The draft mode, at 2K, for two cents a shot
Six 2K drafts for about two and a half full-price renders
Three source photos, and 'Picture 1' is the one that wins
The revision the LoRA people actually train on
Three widgets, one image, and nothing you can tune around it
The API-only Qwen, rented by the picture
The first Qwen node in this pack with a negative prompt
Same schema as 3.0, a third more money, better skin and glass
Qwen Image 3.0 Pro · Text to Image — the sign-painting model, borrowed
Qwen Image 3.0 · Text to Image — the sensible tier of Alibaba's type machine
Qwen Image Edit 2509 LoRA · Edit — your own adapters on Alibaba's editor
Wan 2.1 Image LoRA · Edit — a strength dial, not an instruction box
Wan 2.1 Image LoRA · Text to Image — photoreal stills with an adapter slot
Wan 2.1 LoRA · Image to Video — the one node here where you touch the sampler
Wan 2.1 LoRA · Text to Video — no anchor frame, so the prompt does all the work
Wan 2.1 LoRA · Video Edit — restyle the clip, keep the motion
Wan 2.2 5B LoRA · Image to Video — four widgets, one flat price
Wan 2.2 5B LoRA · Text to Video — the cheapest way to find out if the idea works
Wan 2.2 Animate · Character Animation — drive a photo with someone else's performance
Wan 2.2 Image LoRA · Text to Image — nine LoRAs, two experts, one still
Wan 2.2 · Image to Video — five seconds, two resolutions, nothing to tune
Wan 2.2 LoRA · Image to Video — the expert-split LoRA trick, on a cloud node
Wan 2.2 LoRA · Text to Video — the expensive one, and why it's worth it
Wan 2.2 LoRA · Video Extend — how you actually get past five seconds
Wan 2.2 · Text to Video — the no-fuss five-second clip
Wan 2.5 · Image to Video — 1080p with sound, and no local option at all
Wan 2.5 · Text to Video — 1080p, five aspect ratios, and it hears you
Wan 2.5 · Video Extend — add ten seconds without a visible seam
The cheap Wan you can't download
Cast a character from a clip, cheap
Multi-shot clips with sound, no weights required
Hold your character across the cuts
Type a scene, get multi-shot footage with sound
Thinks first, then hands you four images
A 4K canvas and a nine-image pile-up
4K stills and a colour palette you control
Four drafts a prompt, and a thinking pass
Thirty seconds, synced audio, and an end frame slot
Same thirty seconds, less waiting
Feed it a PDF, a link, or five clips
Thirty seconds, sound included, no weights
Bolt another thirty seconds onto the end
4K is a resize, and that's fine
The most expensive button in the pack
The full reference deck, rendered fast
Prompt-only, at thirty-five credits a second
Another thirty seconds, at the top tier
A document, a link, and five clips
The 4K tier is a 1080p render in a bigger coat
You only pay for the seconds you add
Wan 3.0 reference to video reads your script, not just your prompt
Wan 3.0 text to video is the tier you should be iterating on
Match the resolution or the seam will show
The non-distilled model, now with your weights in it
Your fine-tunes without the 6B download
Two cents a redraw, and it shows in the good way
Trigger words, two cents, go
A thousand photoreal images for about $1.50
Pose and depth control, without the 12GB
Masked inpainting, or just img2img with taste
The classic, with the plastic dial you can turn down
The one specialist inpainting model, now a node
The anti-plastic FLUX, applied to your photo
Photorealism without the plastic sheen
Under a cent a redraw, and that changes your habits
Four fields and a tenth of a cent
Three reference photos, one scene, no 48GB card
The rare Flux 2 fine-tune, running somewhere with the VRAM
Running a FLUX.2 Dev LoRA Without the RTX 6000 Pro
FLUX.2 Dev Is a 53GB Download. This Node Runs It For 1.7 Cents.
FLUX.2 Flex Is For Fixing Type, Not Vibes
FLUX.2 Flex Will Spell Your Headline Right — For About 9 Cents
Klein 4B, Two Cents
Klein 4B LoRA Edit
Your Own LoRA on Klein 4B, Without a 12GB Training Rig
FLUX.2 Klein 4B as a Draft Machine
The Photo Editor People Stopped Arguing About
Klein 9B LoRA Edit
Klein 9B LoRAs Are Where the Ecosystem Lives. Here's How to Stack Them.
The Sane Default of the Flux 2 Line
The Fine-Tune Test Bench Nobody Uses
\"Base\" Is Not the Better Klein — It's the One You Fine-Tune
When Klein Base 9B Is Actually the Right Call
Where Most Klein LoRAs Were Actually Trained — And How to Test Them Natively
FLUX.2 Max Edit Cheaper Than Flex? The Top-Tier Retouch Node
FLUX.2 Max Is 10 Cents an Image. Here's When That's Actually Fair.
8.7 Cents to Change a Photo and Keep the Brand Colour
The Tier You Can't Download, For 4.5 Cents a Shot
The draft tier is the whole point
Three images in, one plain-English instruction
Your own LoRA, on someone else's GPU
A LoRA slot on a model you don't have to run
Stop writing tag soup
It makes the sound too, so write the sound
The most expensive button in this pack
The source clip isn't in the output
The 720p workhorse, with a last frame you can use
Images, clips and audio as source material
And there's a web-search toggle in there
Cheap seconds on the end of your clip
4K, and you'll pay for it
The cheap way to find out if the shot works
Cheapest per second, with an asterisk
Draft video at a draft price
More seconds without starting over
Nine photos, one shot, no GPU
The node that made people angry about pricing
Keep the shot going instead of re-rolling it
30 seconds from one frame, for a price
Up to 30 images, and the billing to match
A whole scene from a sentence
From a 4-second clip to a 30-second take
A whole audio scene, not a voice-over
Ten references and exact pixel sizes
The node with two widgets and one job
Transparent PNGs and ten references, for 4 credits
Pull a finished image apart
Four widgets, and the tiers are free
Same-size 2K outputs, built for a consistent series
Pinned 2K sizes, no surprises
Ten references, transparent output, and a thinking pass
Split a finished frame into usable pieces
The one that rewrites your prompt first
Tell it how to read the line, not just what to say
The cheapest node in the pack, and the most anime
Run your anime LoRA on someone else's GPU
Anime from a prompt plus your own LoRA stack
The cheapest way to find out if Anima is your model
One node, 248 models, and a JSON field you have to get right
Narration and two-hander dialogue, from a node
30 voices, three cents a minute, one catch in the text box
The middle Gemini voice, and the one I'd actually pick
14 reference images, 4K out, and a web search before it draws
The model that spells things correctly
The workhorse edit node, without the 2.1 premium
Mid-price Gemini images, up to 4K
Pay for the one that plans before it draws
Infographics, posters, and text that survives
Get the lyrics out of a sung vocal
A photoreal community mix, on demand
Photoreal faces for 3.6 credits a shot
Animate a still, at 720p or 1080p, for pennies
1080p, silent, and priced like it means it
1080p from a sentence, if you can afford the sentence
1080p clips that arrive with sound
The cheapest Kling clip you can queue in ComfyUI
Make a photo copy someone else's dance
Your still, moving, with sound if you pay for it
The Kling tier where sound becomes optional
Retarget motion and keep the driver's audio
One still, several shots, an optional last frame
Shot lists, 4K, and a prompt the model can cut up
Faster, and it costs more per second
The fast tier, priced like a premium one
30 seconds, and keyframes in between
A casting sheet, a motion reference, and @tags
The most expensive button in the pack
4K and multi-shot without the 4.0 bill
Named subjects with <<<image_1>>> and JSON elements
Kling's unified model, prompt-only, with shot lists
Change one thing and leave the rest alone
Cinematic keyframes for nine cents
Four and a half cents, and a genuinely awkward question
Ten style references for two cents
Bring your own LoRA to a hosted model
Your own style file, somebody else's GPU
Eight reference images and a prompt that has to keep them straight
The thinking step, explained without the hype
A 20-second clip with sound, from one still
Your style, applied to a cloud render
Prompts, trigger words and a soundtrack
Put the soundtrack on its own Audio: line
Keep the clip going instead of starting over
The version with stereo sound and better faces
Three adapters and a stereo soundtrack
Bookend the shot, all the way up to 4K
Twenty seconds, four resolutions, one honest warning about cost
The one wire that lets a cloud model wear your style file
One photo in, a looping turntable shot out
Nine references, three LoRAs, one prompt
The cheap way to test whether your LoRA works
No seed, no LoRA, just a frame and a bill
First frame, last frame, your style in between
Photos, clips and audio as one casting call
480p, three adapters, and a soundtrack in the same pass
Extend a MiniMax H3 clip without the style falling off a cliff
First frame, last frame, done
1 default will bite you: MiniMax H3 Max Text to Video
Add seconds, keep the seam invisible
MiniMax H3 Reference to Video
A look, not an upgrade
Nine sockets and one prompt
Continue a clip in the same style it was made
The cheapest way to get video with sound
Turn a 5-second clip into a scene
A finished stereo song from one node
400 voices, and one field does the acting
Same 400 voices, 40% cheaper per character
1.8 credits an image for a proper anime model
Sixteen references and a mask, for pocket change
The cheap end of OpenAI's image stack
4K masked edits you can actually print
The 4K tier for words in pictures
The auto aspect ratio nobody tells you about
Legible text without the 2.5 price tag
Anime to Real turns your OC into a photo — and quietly redraws the face
One socket, 0.7 credits, and a black background you didn't ask for
Camera Angle Changer re-shoots your scene — as long as it can guess what's off-frame
Video Sound Effects gives your silent clip foley — but it returns audio, not video
9 — you don't get to say what it paints
Image Upscaler goes to 8K for the price of 2K — the catch is what comes back
Name the thing, brush the mask, or do both — both is better
Pose Changer copies a pose from a photo — the text box is only the small print
Relight does the Qwen-Edit relight trick from a sentence — 4.5 credits a go
Shadow Remover lifts the harsh shadow off a face or a product — check the label after
Video Upscaler keeps your audio and your frame rate — mind the 90 MB upload cap
Your Pony tags, someone else's GPU, three of your LoRAs
The cheapest way into Pony, with SDXL bucket sizes instead of ratios
Five source images, one LoRA, 1.8 credits — the cheap multi-image editor
Custom width and height that actually apply
The exaggeration dial survives the trip to the cloud
Same price as v6, faster — so use it to find out if the idea is any good
Two songs per run, and lyrics are optional
Same price as v6, deliberately less predictable
HunyuanVideo 1.5 image-to-video without the 14GB of VRAM
HunyuanVideo 1.5 without buying 14GB of VRAM for it
Vidu Q3 will make your still image act — 16 seconds at a time
A 540p node that costs less than Q3 and more than Turbo
Vidu Q3 Turbo at 720p costs less than plain Vidu Q3 at 540p
Chaining Vidu Q3 Turbo Video Extend in the same graph
4K video with dialogue for the price of a coffee, if you skip the 4K
Cast a character from 15 reference images, no LoRA, no training run
Extending a 4K talking clip without wrecking the voice
Subtitles and transcripts inside ComfyUI, for about four cents an hour
71 voices, one line of text, and a `[laugh]` where you want it
PopNow is a browser-based AI studio for videos, images and music: image-to-video, text-to-video, AI images,
music and speech, one-tap trending video effects and image styles. This pack brings its developer API
(https://api.popnow.ai/api/v1) into ComfyUI graphs. Each model is one node whose inputs come from that
model's published input schema, so the node offers exactly what the model accepts. Generation runs on
PopNow's servers and is paid in US dollars from your prepaid PopNow API balance (no subscription; every run
is quoted first and failed runs are not charged). The API balance is separate from popnow.ai website
credits: add funds in the API console. Make it pop.
344 model nodes (172 video, 139 image, 30 audio, 3 text/vector) from 194 model families, plus PopNow Generate and PopNow LoRA.
<!-- stats:end -->Quickstart
-
Install. In ComfyUI Manager, search for PopNow and click Install, or use the Comfy CLI:
comfy node install popnow-comfyui -
Add your key. Create a key at popnow.ai/api/console/keys, restart ComfyUI, open Settings > PopNow and paste it (or start ComfyUI with
POPNOW_API_KEY=sk-popnow-...set). A notice shows your API balance in USD. Restart once more so the node list comes from your account. Add funds to the API balance at popnow.ai/api/console/billing. -
Add a node and run. Double-click the canvas and type a model name (
seedance,kling,wan,minimax,nano banana,suno...), or browse PopNow > Video / Image / Audio. Wire Load Image into its image input if it takes one, connect Save Image, Save Video or Save Audio, and queue.
Prefer a ready graph? Open ComfyUI's template browser and look under popnow-comfyui: text to image, image to video, text to image to video, text to speech, LoRA, and PopNow Generate.
Features
| | |
| --- | --- |
| What it adds | One node per PopNow model, a PopNow Generate node that takes any model by name, and a PopNow LoRA node |
| Inputs and outputs | Standard ComfyUI IMAGE, MASK, VIDEO, AUDIO and STRING |
| Hardware | None: models run on PopNow, so it works on laptops, Macs and CPU-only machines |
| Dependencies | None beyond ComfyUI itself (Python 3.10+, the V3 node API) |
| Pricing | USD from your PopNow API balance; the rate is shown on every node and each run is quoted first |
| New models | Picked up automatically: the node list is rebuilt from your account's catalog at each start |
| License | MIT |
Supported models
The node list follows the live PopNow catalog (GET /api/v1/models). At the time of this release:
Video (172 nodes)
| Maker | Models | | --- | --- | | PopNow tools | Video Background Remover, Video Face Enhancer, Video Face Swap, Lip Sync, Video Music Remover, Talking Avatar, Video Eraser, Video Restorer, Video Translator, Video Upscaler | | Sand.ai | daVinci-MagiHuman | | Black Forest Labs | FLUX 3 | | Google | Gemini Omni 1.1 Flash, Veo 3.1, Veo 3.1 Fast, Veo 3.1 Lite | | Alibaba | HappyHorse 1.1, Wan 2.1 LoRA, Wan 2.2, Wan 2.2 5B LoRA, Wan 2.2 Animate, Wan 2.2 LoRA, Wan 2.5, Wan 2.6, Wan 2.6 Flash, Wan 3.0, Wan 3.0 Prime, Wan 3.0 Pro, Wan 3.0 Pro Prime | | Tencent | HunyuanVideo 1.5 | | Kuaishou | Kling 2.1, Kling 2.1 Master, Kling 2.5 Turbo, Kling 2.6, Kling 3.0, Kling 3.0 Turbo, Kling 4.0, Kling Avatar 2.0, Kling O3 | | Lightricks | LTX-2 19B, LTX-2 19B LoRA, LTX-2.3 LoRA, LTX 2.5 | | MiniMax | MiniMax H3, MiniMax H3 360 Orbit, MiniMax H3 Character Swap, MiniMax H3 Edit, MiniMax H3 HighSpeed, MiniMax H3 Lite, MiniMax H3 LoRA, MiniMax H3 Max, MiniMax H3 Singularity LoRA, MiniMax H3 X, MiniMax H3 X Turbo | | ByteDance | OmniHuman 2.0, Seedance 2.0, Seedance 2.0 Fast, Seedance 2.0 Mini, Seedance 2.5, SeedVR2 Video | | ComfyUI | OpenJutsu Character Swap, OpenJutsu Object Swap | | PixVerse | PixVerse V6, PixVerse V6 Effects, PixVerse V6 Max | | Topaz Labs | Topaz Astra, Topaz Starlight, Topaz Video AI | | VEED | VEED Dubbing, VEED Dubbing Pro | | Vidu | Vidu Q3, Vidu Q3 Pro, Vidu Q3 Turbo, Vidu Q4 Preview |
Image (139 nodes)
| Maker | Models | | --- | --- | | Alibaba | Qwen Image 2, Qwen Image 1.0 LoRA, Qwen Image 2.1, Qwen Image 2.1 LoRA, Qwen Image 2.1 Pro, Qwen Image 2.1 Turbo, Qwen Image 2512 LoRA, Qwen Image 3.0, Qwen Image 3.0 Pro, Qwen Image Edit 2509 LoRA, Wan 2.1 Image LoRA, Wan 2.2 Image LoRA, Wan 2.7, Wan 2.7 Pro, Z-Image, Z-Image Base LoRA, Z-Image Turbo LoRA | | CircleStone Labs | Anima Turbo, Anima Turbo LoRA | | PopNow tools | Anime to Real, Background Remover, Camera Angle Changer, Face Enhancer, Face Swap Pro, Face Swap, Face Swap Plus, Head Swap, Image Expander, Image Upscaler, Celebrity Look-Alike, Object Eraser, Object Extractor, Photo Colorizer, Old Photo Restorer, Photo to Anime, Pose Changer, Relight, Shadow Remover | | Black Forest Labs | FLUX.1 Dev ControlNet LoRA, FLUX.1 Dev LoRA, FLUX.1 Fill Dev LoRA, FLUX.1 Krea Dev LoRA, FLUX.1 Schnell LoRA, FLUX.2 Dev, FLUX.2 Dev LoRA, FLUX.2 Flex, FLUX.2 Klein 4B, FLUX.2 Klein 4B LoRA, FLUX.2 Klein 9B, FLUX.2 Klein 9B LoRA, FLUX.2 Klein Base 4B LoRA, FLUX.2 Klein Base 9B LoRA, FLUX.2 Max, FLUX.2 Pro, FLUX 3 Klein Base, FLUX 3 Klein Base LoRA | | OpenAI | GPT Image 2, GPT Image 2.5 Flare, GPT Image 2.5 Sunburst | | Ideogram | Ideogram Vector 1.0 | | J1B | Jib Mix Qwen, Jib Mix Qwen LoRA | | Krea | Krea 2 Large, Krea 2 Medium, Krea 2 Medium Turbo, Krea 2 Turbo LoRA, Krea 3 | | MiniMax | MiniMax H3 Image LoRA | | Google | Nano Banana 2, Nano Banana 2.1, Nano Banana Pro | | Neta.art Lab | Neta Lumina | | Pruna AI | P-Image LoRA | | Prefect | Prefect Pony XL, Prefect Pony XL LoRA | | Recraft | Recraft 20B SVG, Recraft V4 Vector | | ByteDance | Seed3D 2.0, Seedream 4.0, Seedream 4.5, Seedream 5.0 Flash, Seedream 5.0 Lite, Seedream 5.0 Pro | | Topaz Labs | Topaz Bloom, Topaz Denoise & Sharpen, Topaz Gigapixel | | Tripo | Tripo H3.1, Tripo P1 |
Audio (30 nodes)
| Maker | Models | | --- | --- | | Resemble AI | Chatterbox | | PopNow tools | Video Sound Effects, Vocal Isolator, Voice Cleaner | | Google | Gemini 2.5 Pro TTS, Gemini 3.8 Flash-Lite TTS, Gemini 3.8 Flash TTS | | OpenAI | GPT-4o Mini Transcribe, GPT-4o Transcribe, GPT-4o Transcribe Diarize, GPT Live Transcribe, GPT-Realtime-2.1, GPT-Realtime-2.1 Mini, GPT Transcribe, TTS-1, TTS-1 HD, Whisper 1 | | xAI | Grok STT, Grok TTS | | HeartMuLa | HeartMuLa Transcribe | | MiniMax | MiniMax Music 3.0, MiniMax Speech 2.8 HD, MiniMax Speech 2.8 Turbo | | Alibaba | Qwen3 TTS | | ByteDance | Seed Audio 1.0, Seed Speech 2.0 | | Suno | Suno v6, Suno v6 Mini, Suno v6 Wild |
Text and vectors (3 nodes)
| Maker | Models | | --- | --- | | OpenAI | text-embedding-3-large, text-embedding-3-small, text-embedding-ada-002 |
Every node with its call name and starting price: MODELS.md.
<!-- models:end -->Install
The pack is on the Comfy Registry as popnow-comfyui. There
is nothing else to install: the nodes only use packages ComfyUI already ships.
ComfyUI Manager. Open Manager > Custom Nodes Manager, search for PopNow, click Install and restart ComfyUI.
Comfy CLI. comfy node install popnow-comfyui
git.
cd ComfyUI/custom_nodes
git clone https://github.com/popnow-ai/popnow-comfyui
# later, to update:
cd popnow-comfyui && git pull
ZIP. Download main.zip, unpack it
into ComfyUI/custom_nodes, and restart.
Used the single-file node from popnow.ai/docs/comfyui before? Delete
custom_nodes/popnow_api.py when you install this pack. Its PopNow Generate node is the same node type
with the same inputs, so saved workflows keep working.
How the nodes work
- Inputs follow the model. Dropdowns, sliders and switches mirror each model's schema. Image, video and
audio inputs are sockets, uploaded for you when the node runs (
popnow://f/...files that only your key can use). Models that take several references grow a new socket each time you connect one. - Defaults are the model's own. A dropdown set to
model default, or a number set to-1, is left out of the request so the model decides. - Outputs are ordinary ComfyUI values. Video nodes return
VIDEO(and the last frame asIMAGEwhenreturn_last_frameis on), audio nodes returnAUDIO, image nodes returnIMAGE, and background removal and layer decomposition also return analphamask. Transcription returnsSTRING. Every node also returns itstask id. - Progress on the node. The node shows the estimate, the queue and generation time, and finally what the run was charged.
PopNow Generate
One node for any model, the same node as the single-file version on popnow.ai:
| Input | What to put in it |
| --- | --- |
| model | A call name from popnow.ai/docs/models, for example minimax/h3/image-to-video |
| prompt | What to make. It goes into prompt (or text for speech models) unless options already sets it |
| options | The model's other fields as JSON, such as {"resolution": "2k"} or {"duration_seconds": 5} |
| image (optional) | An IMAGE, uploaded and placed in the field named by image_field |
| image_field | image_url for most video models, image_urls for image editing (a list; every image in the batch is sent) |
Outputs: image (the result for image models, a blank placeholder otherwise), url (the signed file URL,
valid for about 20 minutes) and path (video and audio are saved to ComfyUI/output/popnow_<task>.mp4 and
similar). Load a saved video back with a video loader node to get its frames.
PopNow LoRA
LoRA models (their names end in LoRA) take a loras input. Add a PopNow LoRA node, set its source
and scale, and connect it; chain several to stack them (most accept up to three). source can be:
| Source | What happens |
| --- | --- |
| A file in ComfyUI/models/loras, such as my_style.safetensors | Uploaded once (up to 2 GiB) and reused for 12 hours |
| A Hugging Face repo id (owner/repo) or page, or a Civitai page | PopNow downloads it once with POST /api/v1/files/import; set HF_TOKEN or CIVITAI_API_KEY in ComfyUI's environment for gated files |
| Any other direct https:// link, or a popnow://f/... URI | Sent as is |
Tokens are read from the environment only, never from a widget: widget values are saved in every workflow and image you share.
Costs and safety
- Price on the node. Each node's badge shows the current rate in US dollars for the options you picked,
for example
$0.037/sfor MiniMax H3 at 768p or$0.048/imagefor Nano Banana 2.1, and updates as you change them.~marks models whose final charge depends on the output. - Quote before every run. Each run calls
jobs/quotefirst, confirms the quoted maximum (maxChargeUsd, sent back asexpectedCostUsd) and holds at most that amount. Set Settings > PopNow > Max USD per run and any run quoted above it stops before anything is held. A limit saved in credits by version 0.1 keeps working (1 credit = $0.001) until you set a USD limit. - Charge on the node. While a run waits, the node shows the estimated cost; when it finishes, the USD
charged (
costUsd). - No surprise re-runs. Seeds start on
fixed, so queueing the same graph twice reuses the first result. Switch control after generate torandomizefor a new result each run (each run is billed). - Safe retries. Every task is submitted with an
Idempotency-Key; network retries reuse it, so a retry never creates a second task. - Your key stays out of shared files. It is stored in the ComfyUI user folder
(
user/popnow/config.json, readable by your account only), never in a workflow or in image metadata. On a server, setPOPNOW_API_KEYinstead. - Cancelling stops the wait, not the job. A cancelled run keeps going on PopNow, is charged if it succeeds, and its result stays in the API console logs.
Configuration
| Setting | Where | Purpose |
| --- | --- | --- |
| POPNOW_API_KEY | environment | API key (sk-popnow-...); takes precedence over the key saved in Settings |
| Settings > PopNow > API key | ComfyUI settings | Saves the key to user/popnow/config.json |
| Settings > PopNow > Max USD per run | ComfyUI settings | Spending ceiling per node run in US dollars (0 = no limit) |
| POPNOW_BASE_URL | environment | API origin, default https://api.popnow.ai (for testing against a mock) |
| POPNOW_COMFYUI_HOME | environment | Folder for the saved key and catalog cache, instead of user/popnow |
| HF_TOKEN, CIVITAI_API_KEY | environment | Used once per LoRA import of a gated file; never stored |
| HTTPS_PROXY | environment | Honoured for every request |
Troubleshooting
| You see | What to do |
| --- | --- |
| No PopNow API key is set | Add the key in Settings > PopNow, or set POPNOW_API_KEY, then run again |
| PopNow 401 | The key is missing, mistyped or revoked; create a new one in the API console |
| PopNow 400 naming a field | A value the model does not accept; check the model page linked in the node's description |
| PopNow 40201 | Not enough API balance: add funds, or lower the resolution or length. Website credits do not pay for API runs |
| PopNow 40202 | The key reached a USD spend cap; raise it in the API console |
| PopNow 40004 | The model cannot serve that combination of settings; change the one named in the message |
| PopNow 40310 | Verify your email address with the link from your sign-up email |
| PopNow content_rejected / generation_failed | The model declined or could not finish; failed runs are not charged |
| An error ending in request_id ... | Send that id to support; it identifies the request exactly |
| a proxy or firewall ... may be intercepting the request | Check that HTTPS_PROXY, if set, points at a working proxy |
| A model from popnow.ai is missing | Restart ComfyUI: the node list is built from your account's catalog at start |
FAQ
How do I generate a video from an image in ComfyUI without a GPU?
Install this pack and add your PopNow key, then add an image-to-video node such as MiniMax H3 · Image to
Video, Kling 3.0 · Image to Video or Seedance 2.5 · Image to Video. Connect Load Image to its
image_url input and Save Video to its video output, write a prompt and queue. The model runs on
PopNow's servers, so your machine only uploads the picture and downloads the clip.
How do I use Kling, Seedance or Wan in ComfyUI?
Each of them is a set of nodes under PopNow > Video (one per task: text to video, image to video, reference to video, video extend). Search the node list for the model name. Resolution, duration, audio and the other options are widgets on the node, with the price for your settings on its badge.
How much does a run cost?
Each model has its own rate in US dollars, shown on the node and in MODELS.md: image models are priced per image, video models per second of output, speech per 1,000 characters or per request. Every run is quoted before it starts, the node prints the final USD charge when it finishes, and failed runs are not charged. Runs are paid from your prepaid API balance, which is separate from popnow.ai website credits. Prices: popnow.ai/api/pricing.
Is my API key safe when I share a workflow or an image?
Yes. ComfyUI writes node settings into workflow files and PNG metadata, so the key is never a node setting.
It lives in a file in your ComfyUI user folder, or in the POPNOW_API_KEY environment variable.
Can I use my own LoRA with a cloud model?
Yes, with the LoRA model nodes. Point a PopNow LoRA node at a .safetensors file in
ComfyUI/models/loras, a Hugging Face repo, a Civitai page or a direct link, and connect it to the model
node's loras input. The lora template comes ready to run with an Apache-2.0 pixel-art LoRA for Z-Image
Turbo (by tarn59 on Hugging Face).
Can I call a model that has no node yet?
Yes. PopNow Generate takes any call name from popnow.ai/docs/models plus its options as JSON. New models also get their own node after a restart once your key is set.
Does it work with ComfyUI Desktop, portable builds or a remote server?
Yes, anywhere ComfyUI can load custom nodes and reach api.popnow.ai. On a server, set POPNOW_API_KEY in
its environment or enter the key in Settings from the browser. Proxies set through HTTPS_PROXY are used.
Links
- Website: popnow.ai · API pricing: popnow.ai/api/pricing
- API console: keys · add funds · Billing: popnow.ai/docs/billing
- Docs: popnow.ai/docs · ComfyUI guide: popnow.ai/docs/comfyui
- API reference: OpenAPI spec · Models: popnow.ai/api/models
- Developer MCP server (API key, USD): popnow.ai/api/mcp · Agent skills: popnow.ai/api/skills
- Other PopNow repositories: popnow-api-mcp, popnow-skills, popnow-python, popnow-js, popnow-api, popnow-blender, n8n-nodes-popnow, popnow-cli, popnow-go, popnow-php
Development
The tests need only aiohttp; ComfyUI, torch and the network are replaced by local stubs and a fake API.
python -m pip install aiohttp
python -m unittest discover -s tests -t .
python scripts/update_snapshot.py --public # refresh the bundled model list from public popnow.ai pages
python scripts/render_docs.py # rewrite the model lists in README.md and MODELS.md
POPNOW_API_KEY=sk-popnow-... python scripts/update_snapshot.py refreshes the snapshot from the
authenticated GET /api/v1/models?includeSchema=1 instead. See CONTRIBUTING.md.
License
MIT © PopNow
3D nodes return mesh download URLs in the file links JSON output. Embedding nodes return vectors and measured usage as JSON in their text output.