PopNow is a browser-based creative studio for video, images, music and voiceovers. One tab, one login, one credit balance, no subscription. You type something or drop in a photo, and a clip, a poster or a song comes back.
That is the pitch. The more useful thing to understand is what sits behind it, because PopNow does not make its own model. It rents other people’s: ByteDance, Kuaishou, Alibaba, Google, OpenAI, Black Forest Labs, Suno and more, all behind one front door and one price list.
What the PopNow AI studio actually is
Start with the catalogue, because the catalogue is the product. The homepage advertises more than 100 apps, 165 AI models on one credit balance and more than 170 one-tap effects. The app directory lists 102 apps grouped by what you are trying to make. The About page still puts it at 154 models from 21 model makers. The site runs in 12 languages and works in a browser on a phone, a tablet or a computer.
There are four ways in: a written prompt, a source photograph, a one-tap effect, or an app from the directory. The effects are prepared prompts with the model and settings already chosen. Pop-Up Book, Cartoon Me Talks, an 80s mall studio portrait, Flying Car Barrel Roll, Ghost Drop, Room Flood, City Fold. Open one and the prompt and model are pre-selected. Swap in your photo, change the wording, generate.
There are also templates built as formats rather than effects: Top 5 Countdown, Mystery Box Unboxing, POV Mini Drama, Text Thread Story, Guess It Quiz. Sign-up is Google, GitHub or an email code, no card needed. That is the point of the front door. You can get to a finished result before you decide whether to pay for anything.
The rest of the toolbox
Video tools cover text-to-video and image-to-video generation, extending a clip, and changing a clip by describing the change in words. Image tools cover generation, photo editing, background removal, object erasing, upscaling, headshots, product imagery and old-photo restoration. Audio covers full songs with vocals, voiceovers and sound effects.
Outside those three columns there is a long tail: anime and comic creation, caricatures, photo collages, memes, GIFs, profile pictures, AI influencers, product advertisements, and free tools for trimming and joining clips. None of it is exotic. All of it is in the same tab, on the same balance, which is the actual argument for a studio over a folder full of bookmarks.
The money, because the money explains the rest
PopNow prices a credit at $0.001. Credits come in one-time packs, from $9 for 9,000 up to $99 for 120,000, and bigger packs carry bonus credits of up to 21%. Do the arithmetic on the top pack. 120,000 credits at the base rate of $0.001 would cost $120, and you pay $99. That is 21% more credits (21,000 bonus on top of 99,000), which works out to $0.83 per 1,000 credits, or roughly 17% off per credit. The discount is real and it is modest.
Every generation button shows its exact credit price before you submit, so you are never guessing at the cost of the thing you are about to make. Failed generations go back to your balance automatically, without you asking. A pack you have not spent any credits from can be refunded within seven days.
The homepage examples run like this: a GPT Image 2.5 Sunburst image from 2 credits, a five-second Seedance 2.0 Mini video from 55 credits, a Suno v6 song from 72 credits. The FAQ gives a slightly different floor for video, 56 credits or $0.056, a one-credit gap that most likely just reflects two pages updated at different times. Those are entry prices, the cheapest end of the range. The pricing page’s own estimates say more about a typical job: the $9 pack covers about 48 five-second videos, 310 images or 125 songs. That implies roughly 185 credits for a typical five-second video and about 29 for a typical image, while songs line up with the 72-credit example. The price moves with the model, the length and the resolution, and longer clips, higher resolutions and newer models all cost more.
What three shots actually cost
Three five-second clips at the homepage’s 55-credit entry rate come to 165 credits, about 16.5 cents of footage. That is the floor. At the pricing page’s typical rate of around 185 credits per clip, the same three shots come to roughly 555 credits, or about 56 cents. Treat both as arithmetic on published figures, not a quote for your job. A different model, a different length or a different resolution will land somewhere else, and the number on the button is the number that matters.
Turn one product photo into a three-shot AI video
The PopNow team wrote a guide on exactly this, using a ceramic travel cup. The method is not about clever prompts. It is about the order you do things in. Three short clips, each with one job: a reveal, a detail, a closing frame. Not one long prompt asked to introduce the product, move to a new location, demonstrate a feature and land on a logo all at once.
Step one: start with a photo you can defend
The shape, the material, the label and the parts that matter should be visible. No hand covering the feature, no packaging in the way, no hard reflection sitting exactly where the video needs to show the surface. And the photo has to be yours, or one you have permission to use. That is not a legal footnote. It is the difference between a video you can publish and one you cannot.
Decide where the video lands before you generate a frame. A vertical social clip and a wide website banner need different compositions. Keep the product away from the edges, where a platform’s interface controls or a later crop will eat it.
Step two: write two lines, not a paragraph
One line for what must stay the same: silhouette, colour, label, number of components. One line for what may change: background, lighting, framing, camera movement. That brief is worth more than a page of style adjectives, because it gives you a test you can apply when the result comes back.
If the lid changed shape, the answer is no, no matter how good the lighting looks. You already wrote down that the lid was not negotiable.
Step three: approve the still frame before anything moves
This is the core of the method. In PopNow’s image workspace you pick an image-editing model that accepts reference images, then describe the composition. For the cup, the guide’s prompt asks the model to keep the shape, glaze colour, lid and visible label, and to place the cup on a pale kitchen counter beside a softly lit window in a vertical frame. No extra products, no added lettering.
Then you compare the still against the original photograph. Lid, handle, seams, label, proportions. If a product detail changed, you fix it now, while you are looking at a single frame. Motion makes a wrong product harder to judge, not easier, because your eye follows the movement instead of checking the shape.
For exact branding the guide is blunt: put the original logo or text over the composition in an editor. Generated lettering should not be your source of truth for packaging or claims. Ads Studio does claim headlines, prices and logos render in brand fonts with crisp text, and the image models are listed at up to 4K with readable text. Even so, a rendered logo is a rendering. Your own file is your own file.
Step four: one action per camera motion prompt
Take the approved still into the video workspace, choose an image-to-video model, and describe what changes over time rather than what already exists in the frame. The guide’s three motion prompts look roughly like this:
Reveal: a slow, steady push toward the cup. The cup stays still. Soft daylight, restrained reflection on the glaze, shape and label consistent throughout.
Detail: a gentle slide from left to right across the rim and lid. Cup stationary, lid closed, minimal motion, no new objects.
Closing: almost still, with a very subtle drift. Keep the clean composition and the empty space beside the product. No added text or objects.
Keep motion small on purpose. A large rotation exposes surfaces the photograph never captured, and the model fills those surfaces in. It will not tell you it guessed. You find out when the label wraps around a curve that does not exist.
Step five: watch each clip to the end and rework one shot
Every clip, start to finish. Not the first frame, not the thumbnail. Watch for a label that changes mid-clip, a lid that opens by itself, a warped edge, a product sliding off its surface. Only the broken shot gets redone. That is the advantage of planning three short clips instead of one long one: the damage is contained to a shot you can re-run.
Then assemble in an editor, cut on stable frames, and drop any section where the product changes. Headline, logo, captions and call to action go on as editable text after the footage is chosen, not baked into the generation. Last check: watch the result at phone size. Can you recognise the product? Can you read the headline without pausing? Does each shot add something the previous one did not?
Why approving the still frame is the whole trick
This is where most AI product video fails, and it fails quietly. Image-to-video models are strong on movement and weak on identity. Hand one a photo and a spin and it will produce something confident and slightly wrong: a drifting label, a lid with an extra seam, a handle that changes thickness halfway through. Nobody notices at thumbnail size. Everybody notices at checkout.
Splitting the job in two turns one hard problem into two easy ones. Composition and product identity get settled in a still, where you can see every pixel and compare it side by side with the original photograph. Motion is added afterwards to an image you already trust.
The diagnostic value is the part people underrate. If the clip goes wrong after you approved the still, you know the still was right and the motion prompt was wrong. That is a solvable problem. Skip the approval step and you have no idea which stage lied to you, so you end up regenerating everything and hoping.
The same logic explains the modest-motion advice. A small push-in or slide keeps the camera inside the evidence the photograph provides. A big rotation leaves the evidence behind and asks the model to imagine the rest.
The AI video and image models you are actually renting
The homepage names Seedance 2.5 from ByteDance, Kling 4.0 from Kuaishou, Wan 3.0 from Alibaba, MiniMax H3, Nano Banana 2.1 from Google, GPT Image 2.5 Sunburst from OpenAI, Seedream 5.0 Pro from ByteDance, the FLUX models from Black Forest Labs and Suno v6. The specifications that decide your workflow:
Seedance 2.5: 1080p clips up to 30 seconds with native sound.
Kling 4.0: up to 30 seconds at 1080p with sound, keyframes, and image, video or audio references.
Wan 3.0: up to 30 seconds with synchronised audio.
MiniMax H3: up to 2K, up to 15 seconds, with control over the first and last frame.
Nano Banana 2.1 and GPT Image 2.5 Sunburst: images up to 4K with readable text.
Seedream 5.0 Pro: high-detail images guided by up to ten reference photographs.
Behind those there is a longer tail: Vidu Q4 Preview, Kling O3, LTX 2.5, Wan 3.0 Prime, FLUX 3, Seedance 2.0 Mini, Nano Banana Pro, Gemini 3.8 Flash TTS, MiniMax Music 3.0, MiniMax Speech 2.8 HD, Qwen Image 2.1 Pro and Krea 3.
Now the awkward part. The counts do not agree with each other. The FAQ says 140 models and 55 image-to-video models. The About page displays 154. The pricing section now says 165. A catalogue that turns over this fast will produce exactly that kind of drift, and the right reading is that any model count is a snapshot of the page on the day someone edited it, this article included.
The studios that sit above the generators
Four things are built on top of the single-model tools, and they are the reason the credit balance is shared rather than split per feature.
Filmmaker
A one-line idea becomes a project with a story, a cast, scenes and a final cut. The cast carries from shot to shot, faces and outfits included. It supports up to 120 shots with the ability to reshoot one shot without rebuilding the film, and up to 100 episodes holding the same cast, so a character change propagates instead of being repeated by hand.
Ads Studio
One product photo becomes a campaign: image ads, product films and creator-style video ads. Six ad types, four output sizes at 9:16, 1:1, 4:5 and 16:9, with no extra cost for the four sizes according to the page. This is the shortest path from the workflow above to something you can actually run.
AI Influencer and Canvas
AI Influencer builds a character from 392 traits covering face, hair, style and setting, keeps that face across posts, and can generate a week of captions and hashtags alongside. Canvas puts images, video and audio on one board, with each result still attached to the card that produced it. Rerun one step and the finished cards upstream are reused rather than charged again, and earlier takes stay in history instead of being overwritten. If you are building a storyboard, that is the feature that saves the most money.
PopNow inside ChatGPT, Claude and the rest
The MCP server is the part most people miss. Connect it once and you can ask for an image, a clip or a song from inside ChatGPT, Claude, Claude Code, Cursor or another MCP client. PopNow says the price appears before a connected job runs and nothing is charged until you approve it.
The GitHub organisation shows how far the plumbing runs: a remote MCP server, agent skills for coding agents, SDKs for Python, JavaScript, Go, PHP and Java, a command-line client, ComfyUI custom nodes, an n8n community node, a server-side proxy for browser apps, a self-hosted Telegram and Discord bot, and a Blender add-on that turns a camera view into renders, video and textures.
For heavier automation, the API accepts reference media as a public HTTPS URL, a Base64 data URI, an uploaded file or an asset-library URI. Uploads take three steps: request a ticket, PUT the bytes to the supplied URL, commit the file. Completed download URLs are signed for about 20 minutes, so results need copying to your own storage or the asset library if you want them to survive the week.
Where it gets thin
Three honest limits, in order of how often they will bite you.
Length is the first. The headline video models top out around 30 seconds per clip, with MiniMax H3 at 15. That is plenty for a product shot and nowhere near enough for a scene. Anything longer is a job for shot assembly in Filmmaker, not a single generation, and planning it as one long prompt will disappoint you.
Generated text is the second. PopNow lists crisp, readable text up to 4K and Ads Studio claims brand fonts rather than garbled lettering. Take the claim seriously and still verify it on your own logo. Type is the first thing to go wrong in a generated image and the last thing a viewer forgives.
The catalogue counts are the third, as above. A wall of model names is a menu, not a guarantee. Run the same still, prompt, duration and output size through two models before deciding which one handles your product, because if you change the inputs between runs you are comparing your own requests rather than the models.
Who it is for, and who it is not
Product marketers with one good photograph and no studio booked. Small teams that need a week of social assets and cannot justify a shoot for six clips. Creators testing a format before committing money to it. Anyone who wants to try a model without subscribing to whichever tool happens to host it this month.
Not for someone who needs a 60-second narrative with a consistent cast across ten scenes, unless they are ready to build it shot by shot in Filmmaker and accept that each shot is a separate decision. Not for a team that needs a signed contract and a seat licence, either. That is a different purchase.
People Also Ask
Who owns the videos and images I make in PopNow?
PopNow states that users own what they make as far as the law allows and that the service claims no ownership. It also says prompts, uploads and results are not used to train AI models. You stay responsible for having the rights to anything you upload and for consent from people shown in it.
What are the upload limits if I connect through the API?
Uploads are capped at 10 MiB for images, and 90 MiB with 600 seconds for both video and audio, with LoRA weights up to 2 GiB. Inline images must be JPEG, PNG, WebP or GIF, no larger than 1 MiB decoded, no more than 8,192 pixels per side, up to 16 per request, with a complete JSON body under 2 MiB.
Can I run one prompt across several models to compare them?
Yes, but the stated ceiling depends on the page you read. The homepage describes comparison of up to eight models, while the About page says up to four. To get a fair read, hold the starting image, prompt, duration and output size constant, otherwise you are comparing different requests rather than different models.
What happens to work I delete?
Creations stay in the asset library until you delete them. Deleted items sit in Trash for 30 days before permanent deletion, which gives you a recovery window if a clip turns out to be better than you thought on second viewing.
The order of operations is the product
None of this is mysterious. A photograph you own, a two-line brief, a still you approve, three small motions, an edit that puts your own logo on last. The models are rented, the price is on the button, and the failed run comes back to your balance.
What PopNow gets right is the sequence. Price before generation. Still before motion. One shot reworked instead of the whole sequence thrown out. None of that is exciting, and none of it is the reason people click on an AI tool. It is, however, the difference between a studio you try once out of curiosity and one you keep open in a tab when a client needs three shots by Thursday.