What building a photo-to-rap video tool taught me about setting expectations

Disclosure: I built AI Rap Video, a browser tool that turns a permissioned adult portrait into a short rap clip with new vocals and music. I used AI assistance to draft this post, then checked every product claim against the live site before saving it. This is a maker write-up with a short walkthrough, not a sales pitch. The tool is independent and is not affiliated with any artist, label, or social platform.

The difficult part of shipping a metered generative feature is not the generate button. It is everything the user needs to know before they press it: what the job will cost, how long the file will stay available, what happens if the run fails, whose photo is allowed, and what the model will not do. If any of that is buried in a FAQ, you get surprised charges, missing downloads, and clips that get shared before anyone has actually watched them.

This post is about the expectation-setting I ended up putting on the form itself. I will also walk through one first run — Original Rap, one person — so those choices are visible in the UI, not just in a list of principles.

Put the cost on the path to the button

AI Rap Video is a browser product. You do not install an app. Generation happens on the homepage; the AI Rap Video Generator page is the map of the same tool. There are two modes. Original Rap builds a new performance from photos, a scene, and a short idea. Video Remix takes a reference clip and recasts it. The walkthrough below stays on Original Rap with a single performer, which is the path with the fewest extra fields.

New accounts receive some free credits, enough to generate a preview video. A full video still requires paid credits. I am not going to invent a starting balance or a single "first run" price here, because the number on the button changes with the model, length, and resolution you pick. What I did instead is make that number visible before submit, and update it as the settings change.

The pricing page publishes one baseline so the pack and plan estimates stay comparable: Seedance 2 Fast, at 12 seconds and 480p, with no video reference, is 45 credits. Raise the resolution, pick a longer clip, switch models, or attach a remix reference, and the quote moves. If the live form and a blog screenshot disagree, believe the form.

That is the first lesson I would pass to another indie maker shipping usage-based AI: put the quote on the path to the button. A billing FAQ after the charge is too late.

Make consent a gate, not a footer

The uploaded portrait is only a reference for who should appear and what they are wearing. The model still invents the performance, which means hands, handheld props, and lip motion can come out wrong. That is a known limit, not a rare glitch, so the finished file has to be watched before it is sent to anyone. It is also why the consent checkbox is a required step, not optional copy under the fold.

The published rules are narrow on purpose:

• Use your own photo, or a photo of another adult who agreed to appear.
• Public figures and minors are not allowed, and neither is anyone who has not agreed.
• Sexual, abusive, or deceptive content is out of bounds.

The photo tips exist because a bad still wastes a run. A sharp, front-facing shot from the head down to the waist, in ordinary light, works better than a cropped selfie or a face lost in a group photo. The accepted types are JPG, PNG, or WebP, each file up to 8 MB. Skip sunglasses, masks, a hand over the face, heavy beauty filters, and duo shots where one person is barely visible.

User uploads do not land in the public example gallery. Completed jobs appear in My videos. That distinction matters: people will not try a tool if they think a private portrait might become a public demo.

A first run, with the expectations left visible

Open the generator, leave Original Rap and One person selected, and sign in when asked. A two-person job is a different form: two portraits, or one shared photo of both adults, plus a required background if you are on the Rap Duo path. Leave that for a later attempt.

Original Rap offers three stages:

• Recording studio, built around an orange booth and a hanging microphone.
• Street cypher, aimed at an outdoor freestyle feel.
• Party entrance, with warmer light and a celebration mood.

The stage chooses the setting. It does not write the lyrics for you. The topic field can stay empty, which produces a freestyle, or you can type a short idea up to 120 characters. The suggestion chips — Birthday, Anniversary, Best friends, Friendly roast, and My business — only drop in an example you can edit. That text steers new lyrics, vocals, and music. It is not a line-by-line lyric editor, and the words are not burned onto the picture as captions.

A concrete note beats a single keyword. A line such as "Mina's birthday — always late, always first on the dance floor" gives the model more to work with than the word "birthday."

Two optional fields are easy to over-read, so the UI has to say what they are not. If you switch to Use my audio as inspiration, you can upload an MP3 or WAV you have the right to use, up to 15 MB. That file can influence timing and direction. It is not used as an unchanged soundtrack, and Original Rap does not promise a recognizable hit or a clone of the person's real voice. A custom backdrop is also optional: same image types, 8 MB limit, and no people in the frame. Skip it, and Original Rap falls back to a plain orange studio background.

Then open Video settings and choose:

• A length of 5, 8, 10, 12, or 15 seconds.
• A frame shape. 9:16 is the usual default for a vertical phone story; 16:9, 1:1, and other ratios are there too. Changing the ratio changes the crop, not the number of people in the cast.
• 480p or 720p.
• A model from the list shown for the current mode. For a first attempt, the default is enough unless you already know you need a different one.

Read the credit quote, confirm the adult-permission statement, and submit.

Say how long the file will be there — and what a failure costs

After submit, status and downloads live in My videos. Pending means the job is still running. When the file is ready, save a copy sooner rather than later: the generated link is kept for up to 7 days. The job record itself remains in My videos after that window.

That retention limit is easy to bury in a help article and painful to discover later. I would rather the form mention it next to the download path. Failures need the same treatment. When a generation is confirmed as failed, the credits for that task return to the balance automatically. If the status is still unclear, it is reviewed first so the same job is not billed twice. Video Remix has an extra case: reference clips that show real people are often stopped by the model's safety check, and those blocked runs are refunded too. Wanting a second take after a completed video means a new job and a new quote.

Play the result with the sound on before you pass it along. Check that the face and clothes still read as the person in the still, that the hands and mouth are good enough for a short fun clip, and that the idea you typed roughly shows up in the performance. If you need to write to [email protected], keep the task ID. Mark the clip as AI-generated when you share it. The product does not post into your chats on your behalf.

Publish the limits next to the feature, not after a complaint

A first run will not give you the person's real voice, and it will not reproduce a trend video frame by frame. It also will not leave a well-known track untouched. Hands, props, and mouth shapes can look off, which is why a birthday clip should be checked before it reaches a family chat.

It is not a full lyric-sheet singer either. There is a separate lyrics tool that drafts 8 or 16 bars of text for 2 credits per set. The video tool still composes its own lines from that short idea. Paste a standout phrase as the topic if you want to steer it, but do not expect a word-for-word vocal.

Credits are shared across the rap scenes, which is another rule that belongs in pricing copy rather than a surprise email. One-time packs do not expire. Monthly plan credits lapse at the end of the billing month. A yearly pool is granted up front and lasts until the end of that billing year. Future renewals can be cancelled in Billing; leftover credits are not cashed out automatically. When you want current pack or plan details, the Pricing page is the live source, not this article.

A short checklist

For a first solo run:

1. A clear portrait of a consenting adult, 8 MB or smaller.
2. Original Rap, one person, a stage, and a topic no longer than 120 characters.
3. Length, frame, and quality chosen, with the credit quote read before you confirm.
4. Collect the file from My videos, watch it with sound, and save it within 7 days.
5. Label the clip as AI-generated, and do not present it as a real performance.

For anyone else shipping a browser AI product like this, the useful part is the same list, just turned around: show the quote before the charge, say how long the artifact will exist, refund a confirmed failure automatically, make consent a gate, and write the limits next to the feature that cannot keep them.

If you follow the walkthrough, the useful report is the browser you used, the settings you chose, the quote the form showed, and what looked different in the file you actually watched. Describing a problem does not require posting the original still.

— Zen Nin

https://airapvideo.cc