AI images, video, audio and 3D for a digital magazine: a worked example

The scenario: a quarterly product magazine

A team of three at a small furniture maker wants a quarterly digital magazine of about 32 pages instead of a printed catalogue. Each issue needs editorial text, product and mood images, a short assembly video, a two-minute audio interview with the designer, a 3D model of one chair and links to the online shop. The team has no video studio and no voice talent, so AI fills the gaps, with a person checking every result. The same steps apply to a report, a brochure or a training manual.

Step 1: Plan the issue and the asset list

Write an asset list before anyone opens an AI tool: for each item, the page, purpose, target format, size or length, reviewer and whether it needs an AI label. This prevents generating a 16:9 clip for a portrait page or an image too small for a spread.

Also decide whether you need an API or a subscription. A team working in a chat app may be better served by a plan from AI subscriptions for work; a team that automates the same steps for every issue pays per use through the API. How to choose an AI model walks through the decision.

Step 2: Text drafts, not finished copy

Any capable text model can draft introductions, product descriptions, headline variants and captions. The quality depends far more on the brief than on the model: give it your style guide, the product facts and one or two past articles as examples. Prompt engineering basics covers prompt structure and examples, and context windows explained shows how much reference material fits into one request.

Treat every draft as a draft. Check product names, dimensions, prices and claims against your own data, because a model can state wrong details with confidence. The editor who approves the text is the one who answers for it.

Step 3: Images

As of October 2026, our catalog lists ten current image generation models (the filter also shows one deprecated xAI model); filter the model directory by image output. Check four things:

• Resolution. Google's Nano Banana models generate at 1K, 2K or 4K, but Nano Banana 2 Lite supports only 1K ^[1]. OpenAI's GPT-Image-2.5 models accept custom sizes with no edge longer than 3,840 pixels and at most 8,294,400 pixels in total, and OpenAI calls sizes above 2560×1440 experimental ^[2]. A 1K image is fine on screen; if the same PDF goes to a printer that asks for 300 ppi, a full A4 page needs about 2,480 × 3,508 pixels, which is more than that pixel cap, so use generated images below full-page size in print or plan an upscaling step.
• Format and transparency. OpenAI outputs PNG by default, with JPEG and WebP as options, and transparent backgrounds in PNG or WebP ^[2]. Cut-out product shots with a transparent background are easier to place in a layout.
• Known weak spots. OpenAI notes that its models can still struggle with precise text placement and with keeping recurring characters or brand elements consistent across images ^[2]. Add headlines and logos in the layout tool instead of asking the model to draw them.
• Real products. Generated images suit mood pictures and backgrounds, not product photos: readers must be able to trust that the chair looks like the picture.

The table shows current list prices for the image models in our catalog.

Multimodal AI explained explains per-image and per-token billing in detail.

Step 4: Video

Video generators turn a prompt or a still image into a short clip. Our catalog does not cover them, so check each provider's documentation for length, resolution and price. One example: Google's Gemini Omni Flash creates videos from text or images and can edit or extend videos; editing or extending uploaded videos is not currently available in the EEA, Switzerland and the UK. It outputs MP4 at 720p by default, with 1080p and 4K as upscaled options, in 16:9 or 9:16, and generates a soundtrack from the prompt. An extension adds 3 to 10 seconds, up to a total of 40 seconds, and every generated video carries a SynthID watermark ^[3].

This area changes quickly: Google's video generation page says the Veo 3.1 Preview models in the Gemini API have been shut down ^[4]. Note the model version in your asset list. For our team, a 30-second mood clip is realistic; for showing how the product is actually assembled, real footage is the honest choice.

Step 5: Audio and voice

Text-to-speech (TTS) models read a script aloud. They are not the speech-to-speech models for live conversations listed in our catalog under audio output. Two documented examples:

• OpenAI lists 13 built-in voices (availability depends on the model), optimised for English, with MP3 as the default output and Opus, AAC, FLAC, WAV and headerless PCM as options. OpenAI requires a clear disclosure to listeners that the voice they hear is AI-generated and not a human voice. The page also lists all three current speech models as deprecated, with shutdown on January 6, 2027; OpenAI points to gpt-realtime-2.1-mini on the Realtime API instead, which the speech endpoint does not support, so it is not a drop-in replacement ^[5]. A magazine that runs into 2027 should plan for another speech model or provider.
• Google's Gemini TTS models offer 30 prebuilt voices, more than 130 languages on Gemini 3.8 Flash TTS, and up to two speakers in one request with prebuilt voices. A normal request returns a WAV file (24 kHz, mono, 16-bit); the two Gemini 3.8 models can also design or replicate voices ^[6].

For the interview, record the designer's real voice. Use a generated voice for product explainers or translations, clone a real person's voice only with their written consent, and keep the script: it becomes the transcript in Step 9.

Step 6: 3D models

Text-to-3D and image-to-3D tools create a mesh from a prompt or a photo. Our catalog does not cover them, so judge each tool by what it exports. glTF is a royalty-free format for 3D scenes and models that can be stored as one binary .glb file ^[7], the simplest format to hand on. For a real product, your own CAD files are usually more accurate than a generated mesh; check any model against the product from every angle.

Step 7: Lay out the issue and export the PDF

Bring text and images together in your layout tool (InDesign, Affinity, PowerPoint, Word or similar) and export a PDF. Leave space for the videos, the audio button and the 3D model, ideally with a still image as a placeholder, and set headlines, logos and prices as real text. Give each meaningful image a short text alternative and mark decorative images as decorative; W3C's tutorial explains how to tell them apart ^[8].

Step 8: Publish the issue with Viewnamic

Here the separate assets become one interactive issue. Viewnamic publishes the PDF you already have in a browser viewer without changing the file; you mark an area on a page and add a video, an audio file, a 3D model, a click dummy, an image slideshow or a link ^[9]:

• Video. MP4, MOV, AVI, WebM or MKV, converted to MP4 in the browser. Clips can autoplay when visible, loop, play with or without sound and hide the controls; Viewnamic can place a video over a picture in the layout ^[9].
• Audio. MP3, M4A, WAV, OGG, FLAC or AIFF, or the soundtrack of a video. It appears as a speaker button and can repeat or keep playing while readers turn pages ^[9]. A WAV file from a TTS model fits directly; headerless PCM needs converting first.
• 3D. GLB, GLTF, OBJ or STL; readers rotate and zoom with mouse, trackpad or touch ^[9].
• Images and links. Slideshows show several images in one area; click dummies are JPG or PNG screens with hotspots ^[9], so export generated variants as PNG or JPEG rather than WebP if you want to use them in a click dummy. Link areas, for example to the shop, are included on every plan ^[10].

Media items count against a plan total: 3 on Single, 10 on High Five, none on Publisher, 50 on Publisher Plus and unlimited on Premium. Video plays in Economy, Standard or Premium quality depending on the plan, which matters if you generated clips at 1080p or 4K ^[10].

Readers need no app, plugin or account. Scroll and fullscreen presentation modes are on every plan; page-flip magazine mode starts with Publisher ^{[11] [10]}. Every publication also has a reading mode, a text version on the same address, including the embedded media, for screen readers, small screens, search engines and AI assistants; indexing is set per publication ^{[11] [12]}.

Embedding on your own website works from High Five, with media working inside the embed. Reader statistics, including media plays, need Premium or Tailored; they use no cookies and store no IP addresses ^{[10] [13]}. The details are on the pages about adding video, audio and 3D models to a PDF and sharing and embedding publications.

Limits: media play only in the online viewer, and the PDF readers can download (on plans that include downloads, if you allow it) stays static ^{[9] [10]}. Viewnamic is for business customers only, and public links are deactivated when the trial or subscription ends ^[14]. Viewnamic's feature pages describe the reading mode as an accessible text version but state no WCAG conformance level ^[11]. You can test a PDF of up to 200 pages with three media items without signing up; the preview is private and deleted after 24 hours ^[15].

Step 9: Review, accessibility and labels

Review. Check the issue on a phone and a laptop, and watch and listen to every generated clip to the end.

Accessibility. W3C recommends a transcript, a text version of the speech and non-speech audio, for audio content; video also needs captions and, where visual information matters, audio description ^[16]. Put the interview transcript on the page or link to it, and use the TTS script as the transcript of a narrated explainer.

Labels. Article 50 of the EU AI Act requires deployers to disclose AI-generated or manipulated image, audio or video content that constitutes a deep fake, and AI-generated text published to inform the public on matters of public interest unless it has undergone human review or editorial control under someone's editorial responsibility. The information must be clear and distinguishable at first exposure and meet accessibility requirements ^[17]. Whether a duty applies to your issue is a legal question; a provider's terms can also require a label, as OpenAI's do for TTS voices ^[5]. A visible caption such as "Voice generated with AI" next to the speaker button is easy.

Rights and provenance. Check each tool's terms for commercial use before you generate, and keep a log of prompts, tools and versions. Watermarks such as Google's SynthID ^[1] and the C2PA standard for content provenance ^[18] help others recognise generated media, but they do not replace a visible label. Our sister site's guide to AI for magazines covers labels, copyright and provenance in more depth.

What the issue costs

A rough estimate for four issues a year, using list prices as of October 2026:

• Text. Assume 200,000 input and 300,000 output tokens per issue for drafts and revisions. Across the base list prices in our catalog, that is between about $0.05 and $17 per issue. Enter your own volume in the AI cost calculator or compare rates on AI pricing.
• Images. 30 images per issue, with three attempts each, makes 360 generations a year: about $3.60 at Muse Image's $0.01 (preview), $12 at Nano Banana 2.1's $0.0336 for 1K, or $48 at Nano Banana Pro's $0.134. Higher resolutions cost more.
• Video, voice and 3D. Not in our database. Check each provider's pricing page, since the billing units differ from provider to provider.
• Publishing. In Viewnamic, a quarterly issue with five media items needs 20 media placements a year, more as the archive grows. High Five (€99 a year) includes 5 PDFs and 10 media items, without page-flip. Publisher (€299 a year) has page-flip but no media. Publisher Plus (€699 a year, or €75.73 a month billed monthly) combines page-flip with 50 media items and reader downloads, enough for this scenario for about two years ^[10]. Prices are net, plus VAT ^[14]. See the Viewnamic pricing and plan limits for the full table.

The largest cost is usually people: the editor who checks facts, the designer who fixes the layout and whoever reviews every generated clip. Plan that time first.

Checklist before you publish

• Every asset exists in the agreed format, size and length.
• A person has checked facts, product details and prices in AI drafts.
• Images have text alternatives, audio a transcript, video captions.
• AI labels are visible where the law, a provider's terms or your rules require them.
• The prompt and tool log is saved, and the issue is tested on phone and desktop.