AI made making things nearly infinite. Text, images, and now video, voice, and characters — one person can now produce them in minutes.
The catch: as supply exploded, so did the number of tools. Seedance, Kling, Veo, Runway, Sora, ComfyUI, Live2D, VOICEVOX, ElevenLabs, HeyGen… you hear the names, but there's no single place that answers the real questions: which one does what, how far can I get for free, can I use it commercially, and can I run it locally?
This article is that map. Six categories of the tools that matter, sorted by what they do / price / commercial use / availability / local-only, written for people making short videos, explainers, and character content solo. Everything is current as of July 2026, with the load-bearing claims checked against primary sources.
Five things to know before the tables
Before the details, five cross-cutting calls. These alone settle most decisions.
1. Don't build on Sora anymore. OpenAI's Sora 2 / Sora 2 Pro shut down its web and app versions on April 26, 2026, and the developer API is scheduled to shut down on September 24, 2026 (it still runs for now, but don't start new pipelines on it). If your generation flow depends on Sora, migrate to Veo, Seedance, or Kling.
2. The workhorses for AI video are Chinese models right now. Seedance (ByteDance), Kling (Kuaishou), Hailuo (MiniMax), Vidu, and Wan (Alibaba) combine "cheap, available outside the US, audio included, lip-sync included." Google Veo and Runway are worth it for the hard shots but cost more. Use a Chinese model as your workhorse and a Western one for the finishing cut.
3. Images: a local + cloud combo is the standard. Character base art and large batches are strongest locally (ComfyUI + SDXL-family models); expression variants and re-staging a character into new scenes are faster and cheaper with editing models like Nano Banana or FLUX Kontext. The era of "train a LoRA just for character consistency" is fading.
4. Japanese-language narration can be produced for essentially zero. Character voices from VOICEVOX, calm narration from AivisSpeech (a local TTS built on Style-Bert-VITS2). Those two cover most of it. Pay for ElevenLabs / OpenAI TTS / Suno only when you need multilingual reach, a distinct voice identity, music, or dubbing.
5. "Free to download" ≠ "free to use commercially." This is where people get burned. With image models especially: NoobAI XL forbids commercial use including of its outputs, and FLUX.1 dev is non-commercial (commercial use requires a separate paid license). Read the license box on the model page every time. More below.
CH1. AI video generation models
The foundation for turning text or an image into video — sequential cuts for vertical shorts, product-style shots, and i2v (animating a still). In 2026, models with native audio and lip-sync lead.
| Tool | What it does | Price (approx.) | Notes |
|---|---|---|---|
| Seedance 2.0 (ByteDance) | T2V, I2V, multi-shot. Generates audio + lip-sync in one pass | ~$0.7 per 5s at 1080p | Excellent image→video and sequential cuts. Available via CapCut / Dreamina; not offered in the US |
| Kling 3.0 (Kuaishou) | Unified T2V/I2V/editing. Native 4K, 60fps, lip-sync incl. Japanese | Free tier → subscription / per-second API | Top of the video-generation arena. Strong physics and anatomy |
| Veo 3.1 (Google) | T2V, I2V, audio in one pass, best prompt adherence | Fast ≈ $0.15/s | Best overall quality — but an invisible SynthID watermark is applied to every output and can't be removed |
| Runway Gen-4.5 + Act-Two | Character consistency from one reference, camera control, face-motion transfer (v2v) | Credit-based / API | Deep editing workflow; good for style transfer |
| Hailuo 02 (MiniMax) | I2V-focused, stable motion and prompt adherence | from $0.017/s at 512P | Ultra-low-cost; good for draft volume |
| Vidu (ShengShu) | Reference-to-Video: 3–7 reference images for a consistent character, up to 16s | from ~$0.03/s | Cheapest path to a consistent recurring character |
| Luma / Pika | 1080p + HDR / 9:16 vertical + morph effects | subscription + usage | Luma for polish, Pika for effects and short formats |
| Sora 2 / Pro (OpenAI) | High quality, but being retired | — | ⚠️ App shut down; API ends 2026/9/24. Don't adopt |
| Open models: Wan 2.5 / LTX / HunyuanVideo / CogVideoX | Local generation with audio (Wan/LTX), prompt-faithful long clips (CogVideoX) | GPU only (no per-generation cost) | No watermark, near-zero marginal cost. Run inside ComfyUI |
Quick picks: best quality → Veo 3.1 · image→video → Seedance · consistent character at scale → Vidu · mass drafts → Hailuo · zero-cost in-house → Wan 2.5 (local).
CH2. 3D & real-time engines
The base for moving 3D characters and rendering cinematic camera work. The "free below $X revenue / indie" tiers matter for solo creators.
| Tool | What it does | Price | Notes |
|---|---|---|---|
| Unreal Engine 5.6 + MetaHuman | High-fidelity digital humans; facial animation from audio or a webcam; cinematic camera, real-time render | Free under $1M revenue | The high-quality trump card. Steep learning curve |
| Unity 6 (Timeline / URP) | Load VRM avatars → animate camera → output to OBS. Lightweight 3D VTuber | Free under $200K revenue | Streaming-oriented; URP is enough for video |
| Blender | Modeling, animation, rendering, 2D animation (Grease Pencil), video editing (VSE) | Fully free, commercial OK | The all-purpose free base. Have this one |
CH3. 2D / VTuber character animation
Making a single illustration talk and move — the core of VTuber-style shorts.
| Tool | What it does | Price | Notes |
|---|---|---|---|
| Live2D Cubism | Rig a layered illustration with bones and physics | FREE tier / PRO $15/mo; indie tier under ~$67K revenue | The industry standard for 2D VTubers |
| Adobe Character Animator | Drive a character in real time with a webcam + mic, auto lip-sync | Starter free / Pro via CC | Improvise a talking explainer while reading a script |
| VTube Studio | Face-track a Live2D model and output to OBS | Free | The most popular, lightest 2D capture tool. Start here |
| VSeeFace / Warudo | Face + hand tracking for VRM (3D) avatars | Free | 3D VTuber capture and scene building |
| VRoid Studio | Create VRM-format 3D characters | Free | Entry point for 3D characters |
CH4. Voice, music & narration
Narration, character voices, music, SFX, cleanup. Japanese narration is essentially free and local. (For non-Japanese narration, the cloud options below carry more of the weight.)
| Tool | What it does | Price | Notes |
|---|---|---|---|
| VOICEVOX | Local Japanese TTS (character voices), scriptable via API | Free | The character-voice standard. Commercial use OK with credit line (check per-character terms) |
| AivisSpeech | High-quality local Japanese TTS (Style-Bert-VITS2 based); emotion/tone control | App free | Best-quality local TTS tier; the pick for calm narration |
| ElevenLabs v3 | TTS, voice cloning, multilingual dubbing, SFX, music in one platform | $5+ (commercial incl.) | Top-tier expressiveness and voice identity. Free tier is non-commercial |
| OpenAI TTS (gpt-4o-mini-tts) | Low-latency TTS you can instruct on delivery | ≈$0.015/min (cheapest tier) | Multilingual, cheapest cloud |
| Suno v5 | Songs / BGM from lyrics + genre | Pro $10 (commercial rights) | Leads on vocal quality and speed. Free is non-commercial |
| Stable Audio | Non-vocal SFX, score, ambient | $12+/mo | Clearest licensing |
| Adobe Podcast Enhance → Auphonic | AI cleanup (noise/reverb) → loudness normalization | free tiers | The standard flow to take mic audio to broadcast quality |
| Applio / RVC (voice changers) | Convert one voice into another, real-time capable | Free OSS | Cloning your own voice is safe; cloning others without consent is not |
Quick picks: character voice → VOICEVOX · calm narration → AivisSpeech · expressiveness → ElevenLabs · cheapest cloud → OpenAI TTS · music → Suno · cleanup → Adobe Enhance → Auphonic.
For what it's worth, this article is written by a small studio that ships apps solo, leaning hard on AI. The voice/image/video pipelines above are the same ones we built to produce our own app trailers and social shorts.
CH5. AI image generation & local pipelines
Making the raw material for video — character art, backgrounds, thumbnails, source stills for i2v. As in point 3, a local + cloud combo is the standard.
| Tool | What it does | License / price | Notes |
|---|---|---|---|
| ComfyUI | The node-based core. Handles images and video (i2v); can be turned into an API | Free OSS | Keep it as the "engine" behind a form/preset UI |
| SwarmUI / Forge / Krita AI | Simple UIs / wrappers over ComfyUI | Free | For "I just want a generate button." Krita AI integrates Kontext editing |
| Illustrious / Animagine / Pony (SDXL family) | The main battleground for anime character art; LoRA/ControlNet for control | Free. Commercial use of generated images: OK | But hosting the model itself as a paid service is restricted |
| NoobAI XL | Illustrious-based anime model | ⚠️ Commercial use forbidden, outputs included | Don't use for monetized video |
| FLUX.1 schnell | High-quality T2I, Apache 2.0 | Free, commercial OK | The clean-license FLUX. Use this for commercial work |
| FLUX.1 dev | Top-quality T2I | ⚠️ Non-commercial (paid license for commercial) | Use schnell or the paid tier for commercial |
| FLUX.1 Kontext | Edit an image with natural language; keep the face, change the scene | Local / cloud | A main method for character consistency |
| Nano Banana 2 / Pro (Google Gemini) | Character consistency from reference images (no training), up to 4K, conversational editing | Usage-based | The "magic" for consistency. Cents per character vs. dollars to train a LoRA |
| Seedream 4.0 (ByteDance) | 4K image gen + editing, reference consistency, cheapest tier | ≈$0.03/image | Ideal for i2v source stills, thumbnails, character sheets |
| Midjourney v7 / v8.1 | Mood boards, photorealism, consistency (Omni Reference) | $10+/mo | For thumbnails/source art; zero local control |
| Wan 2.2 (local i2v) | i2v inside ComfyUI; runs on a 12GB GPU with GGUF quantization | Free | Draft volume at near-zero cost; pass finishing cuts to an external API |
CH6. Talking avatars, 3D generation & motion
Script → talking video, image → 3D model, single-camera motion capture. These collapse the prep time for explainers, multilingual reach, and 3D assets.
| Tool | What it does | Price | Notes |
|---|---|---|---|
| HeyGen | Script → talking video. 175-language dubbing, gestures, micro-expressions | Free tier → $29/mo | Top overall; the all-rounder for multilingual explainers |
| Hedra (Character-3) | Still → talking character, natural for social, 140 languages | Free → $75/mo | For running a recurring character |
| Captions (Mirage Avatar X) | One-shot vertical shorts, auto captions, an AI twin | Free → $29.99/mo | Vertical-short specialist |
| D-ID | Make one face photo talk | $5.99/mo+ | Cheapest option |
| Meshy / Tripo / Rodin / Luma | Image/text → 3D model, real-object scanning | Free tiers + subscription | Meshy = all-round, Tripo = fast/free, Rodin = quality, Luma = scanning. Finish in Blender |
| Rokoko Vision / Move.ai | Markerless mocap from a webcam | Rokoko single-cam free | Try Rokoko first; use Move.ai when you need quality |
Quick reverse-lookup guide
Starting from "what I want to make," this is the shortest path.
| Goal | Start with | Reinforce / alternative |
|---|---|---|
| Vertical character shorts | VOICEVOX + Live2D/VTube Studio + Wan 2.2 (i2v) | AivisSpeech for narration; Seedance for finishing cuts |
| Mass-produce one consistent character | Illustrious+LoRA → Nano Banana 2 → Vidu | base art → expression variants → video |
| Faceless explainer videos | HeyGen / Captions (script → talking avatar) | expand straight into multilingual shorts |
| Product comparison / product-style shots | Seedance multi-shot | AivisSpeech for narration |
| BGM & SFX | Suno (songs/BGM) / Stable Audio (SFX) | Generate only the sounds you can't find free |
| Everything free & local | ComfyUI + Illustrious/FLUX schnell + Wan 2.2 + VOICEVOX | A 4090-class GPU (16–24GB) + 32GB RAM is comfortable |
The licensing traps (please read this part)
The most powerful free models tend to hide the catch in their commercial terms. Check before you monetize.
- NoobAI XL — the model, its derivatives, and its outputs are all forbidden for commercial use. Free and high-quality, but not usable for monetized video.
- FLUX.1 dev — non-commercial. Commercial use requires a paid license from Black Forest Labs. For commercial work, use FLUX.1 schnell (Apache 2.0).
- Illustrious / Animagine / Pony family — using the generated images commercially is fine. But the license (Fair AI Public License family) separately restricts hosting the model itself as a paid inference service. Selling images: OK. Standing up a paid API on the model: a different question.
- Veo 3.1 — every output carries an invisible SynthID watermark that survives compression and cropping. For work where a watermark is unacceptable, use Seedance / Kling / open models.
- VOICEVOX — commercial use is OK with a credit line (e.g. "VOICEVOX: Zundamon"), but each character has its own terms, so check the ones you use.
One rule: "free to download" and "cleared for commercial use" are different things. Make reading the license box a habit.
Who's writing this
Particles and Waves is a small studio that ships apps solo, leaning hard on AI. The voice, image, and video workflows above are the ones we built to get our own apps to a global audience. If you're on the "maker" side too, we hope it's useful. A few of the things we make:
- notaps Words — a vocabulary app that grows your words through passive listening, writing, and pronunciation practice (20 languages, no ads).
- notaps Oshi — auto-collects updates about the niche creators and idols you follow.
- GoodsCri — turns a single favorite photo into a glowing, animated piece of fan merch.
Current as of July 2026. Prices are approximate and include third-party aggregation. Verify each service's official page and license before adopting or producing at scale. The only retirement confirmed by an official source is OpenAI's Sora.