Bài thử thật các công cụ AI giọng nói, ảnh và video của mọi hãng, gồm ElevenLabs, Higgsfield, Synthesia, GPT Image, video và giọng đọc của Gemini, Voxtral. Mỗi bài ghi kết quả và chi phí thật, để bạn so các công cụ cạnh nhau.
2 bài
ElevenLabs
Đo thật API tạo tiếng động ElevenLabs bằng gói Free: 10 credit mỗi giây khi đặt độ dài, 50 credit mỗi clip khi để máy tự chọn, chỉ bằng 1/4 giá ghi trong tài liệu. Đề tiếng Việt ra gần như im lặng mà vẫn bị trừ credit, kèm giá quy ra đồng.
Higgsfield
Cách nối Higgsfield MCP vào Claude Code và Claude, kèm số tiền thật: ba lần tạo trên gói Plus tốn 5,5 credit (khoảng 6.950 đ), báo giá khớp số bị trừ, prompt tiếng Việt ra chữ đúng dấu, còn Unlimited bị từ chối trong Claude Code.
Chủ đề này chưa đủ bài tiếng Việt, nên kèm thêm các bài tiếng Anh cùng chủ đề; bấm vào sẽ mở bản đó.
ENElevenLabs
We made 25 sounds with the ElevenLabs Sound Effects API on a Free plan key and read the credit counter after each call: 10 credits a second at fixed length, a flat 50 on auto length, a quarter of what the docs say. Vietnamese prompts came back almost silent.
ENHiggsfield
How to connect Higgsfield MCP to Claude Code and Claude, and what it really costs: we ran three jobs on a paid Plus plan, read the ledger, and found every quote matched the charge, while the unlimited offers were refused in Claude Code.
ENHiggsfield
We animated one product photo with six image-to-video models on Higgsfield's Plus plan and read the label in 12 frames of every clip. Clips cost 5 to 12.5 credits; Seedance 2.0 Fast changed a letter on one of two runs and Kling 4.0 reshaped the bag.
ENChatGPT
We made 10 copies of one text-heavy banner with GPT Image 2.6 on Higgsfield's Plus plan and read every charge. 4K cost 1.25 credits for 11.6 times the pixels of the 0.5-credit default, and high quality cost six times low for the same text.
ENGemini
We generated three dialogue clips with Gemini Omni 1.2 Flash over the API. A 15-second 720p clip cost $1.05 and took under 90 seconds, every scripted line came out word for word in English and Vietnamese, but 360p gave a different scene and one clip added a TV logo.
ENSynthesia
We made two avatar videos on Synthesia's free Basic plan. Both rendered in 1080p in about seven minutes, both carried a Synthesia logo, and Download opened a STARTER+ upgrade box instead of an MP4.
ENChatGPT
We sent the same text-heavy banner prompt to both GPT Image 2.6 models over the API. Both spelled every line right, even in Vietnamese, at $0.034 an image; Flare returned in about 9 seconds, Sunburst in 21, and Sunburst added text we never asked for.
ENElevenLabs
We sent one tricky English paragraph through five ElevenLabs and Gemini voice models over the API, three times each. Eleven v4, v4 Turbo and Gemini 3.8 Flash TTS read back with zero errors; Gemini cost $0.0136 a minute, Eleven v4 $0.0624.
ENHiggsfield
We sent one prompt to seven Higgsfield video models and read the charges in the ledger. A 5-second clip cost 5 to 35 credits, the cheapest model passed twice, and unused plan credits expire. Credits per video, cost per plan and what 1,000 credits really buy.
ENElevenLabs
We timed the audio ElevenLabs returned for each credit on a free API key: Flash v2.5 gave about 125 seconds per 1,000 credits, v4 and v3 about 62–63. What Free, Starter, Creator and Pro list in minutes, and why the API page shows far more than the web app.
ENElevenLabs
We created a free ElevenLabs API key with a 2,000-credit cap and ran Flash v2.5, Multilingual v2, v3 and v4 in English and Vietnamese: what each call cost, how fast it was, which models read Vietnamese correctly, and what the free tier blocks.
ENMistral
We scored Mistral's Voxtral Mini Transcribe 3, whisper-1 and gpt-4o-transcribe against the script of two narrations. English: 1.6%, 1.7% and 2.1% word error rate. Vietnamese: 1.7% against 6.8% and 6.6%, with 37 of 39 listed names and terms right against 16 and 21. Voxtral also lists at USD 0.0035 a minute against 0.006.
