Seedance 1.0 Pro Fast

Seedance 1.0 Pro Fast — the fast, cheap Seedance 1.0 Pro; 720p costs just 7 credits per second.

Seedance 1.0 Pro

Seedance 1.0 Pro — ByteDance's steady video generation, 480p to 1080p, up to 12 seconds.

Seedance 2.5 (API)

Seedance 2.5 on the official BytePlus API — 480p to 1080p, 4–15 seconds, with reference image/video/audio input.

Wan2.2 I2V Flash (Ảnh→Video 480P)

Wan2.2 I2V Flash — a silent 5-second clip from one image; 480P costs just 25 credits. The cheapest lane for testing ideas.

Wan2.6 I2V Flash (Ảnh→Video, có tiếng)

Wan2.6 I2V Flash — the budget Wan image-to-video: pick silent or with audio, 2–15 seconds. Silent 720P is just 8 credits per second.

Wan2.2 KF2V Flash (Frame đầu → frame cuối)

Wan2.2 KF2V Flash — give a first and last image and the model builds the transition between them. Output is always 5 seconds.

Wan2.2 Animate Mix (Thay nhân vật trong video)

Wan2.2 Animate Mix — swaps the character in an existing video for the one in your image, keeping the original motion and scene. Billed by output length.

Wan2.2 Animate Move (Nhân vật nhại động tác)

Wan2.2 Animate Move — give one character image plus a driving video; the character mimics the motion. Billed by the output video length.

Vidu Q2 Pro Fast (Ảnh→Video)

Vidu Q2 Pro Fast — the cheapest, fastest video lane available: one image to a 720P/1080P clip at 5 credits per second.

Vidu Q3 Drama (Phim / kịch)

Vidu Q3 Drama — the film/drama variant: acting, character emotion and cinematic cuts; 1080P costs the same as 720P.

Vidu Q3 Ad (Video quảng cáo)

Vidu Q3 Ad — the product-advertising variant: keeps the exact shape, label and material of the product in your reference images.

Vidu Q3 Mix (Tham chiếu→Video)

Vidu Q3 Mix — feed 1–7 reference images (characters, props, scenes) to build a video that keeps their identity. Up to 16 seconds.

CinAI Gemini Omni

1 free runs/mo

CinAI Gemini Omni — text-to-video with audio and a thinking step. Supports image-to-video (start image), 16:9 / 9:16 aspect ratios, up to 10 seconds.

Omni 1.1 Flash

10 free runs/mo

Omni 1.1 Flash (Google Labs Flow) — video with audio from text, or referencing 1–4 character/object images and one video clip. 16:9 / 9:16, 4/6/8/10 second duration. Async.

Grok Video 1.5

5 free runs/mo

Generate video via grok.com/imagine (browser, requires SuperGrok). Text-to-video (prompt only) or image-to-video (with 1 image). Aspect Ratio applies to text-to-video; with image input (i2v), the original image's aspect ratio is kept.

Wan3.0 Video Prime

Wan3.0 Video Prime — the premium Wan3.0 with more lifelike visuals and consistent characters, videos up to 30 seconds with sound.

Wan3.0 Video

Wan3.0 Video — Alibaba's newest all-in-one video model: text-to-video, first/last-frame and reference-image video up to 30 seconds, with sound.

Seedance 2.5

Text-to-video & image-to-video via Dreamina Seedance 2.5 (ByteDance) — the LATEST version, highest quality (Dola: "For best video quality"). Pass a prompt (t2v) or add reference images (i2v, up to 10 images). Duration 5/10s, 6 aspect ratios. Async: 5s ~1-3 minutes, 10s ~5 minutes.

Seedance 2.0

Text-to-video & image-to-video via Dreamina Seedance 2.0 (ByteDance) — browser automation. Pass a prompt (t2v) or add a reference image (i2v). Duration 5/10s, multiple aspect ratios. Async 1-3 minutes.

CinAI Veo 3.1

CinAI Veo 3.1 — CinAI's highest-quality video model. Text-to-video and image-to-video (image ≤20MB, first + last frame). 4/6/8 second duration, 720p/1080p, 16:9/9:16, 24 FPS, with audio. Async (~60-90s).

CinAI Veo 3.1 Fast

1 free runs/mo

CinAI Veo 3.1 Fast — fast, cost-effective video generation. Text-to-video and image-to-video (image ≤20MB, first + last frame). 4/6/8 second duration, 720p/1080p, 16:9/9:16, 24 FPS, with audio. Async (~40-70s).

Dreamina-Seedance-2.0-mini

Multimodality-to-video generation model by ByteDance/Doubao: takes text + image + video + audio references → video (may include audio). 480P/720P, 4-15s, 24fps.

Dreamina-Seedance-2.0-fast

Seedance 2.0 Fast by ByteDance/Doubao — faster multimodal VIDEO generation: text + images (1-9) + video (0-3) + audio (0-3) references → 1 video (may include audio). 480P/720P, 4-15s, 24fps; modes include Text→Video, Image→Video, First/Last Frame, Multimodal Reference.

Dreamina-Seedance-2.0

Seedance 2.0 (full version) by ByteDance/Doubao — generates multimodal VIDEO: text + image (1-9) + video (0-3) + audio (0-3) references → 1 video (which may include audio). Supports 480P/720P/1080P/4K, 4-15s, 24fps; modes include Text→Video, Image→Video, First/Last Frame, and Multimodal Reference.

Veo 3.1 Lite

10 free runs/mo

Veo 3.1 Lite (Google Labs Flow) — lightweight Veo 3.1 text-to-video with audio. 16:9 / 9:16, 4/6/8 second duration. Async.

CinAI Veo 3.1 Lite

2 free runs/mo

CinAI Veo 3.1 Lite — the most affordable model in the CinAI Veo line (Preview). Text-to-video and image-to-video (first frame). 4/6/8 second duration, 720p/1080p, 16:9/9:16, 24 FPS, with audio. Async (~40-70s).

HappyHorse 1.1 (Text→Video)

HappyHorse 1.1 text-to-video by Alibaba (Tongyi) — generates video from a PROMPT. 720P/1080P, ratios 16:9/9:16/1:1/4:3/3:4, up to ~10s. DashScope async.

HappyHorse 1.1 (Image→Video)

HappyHorse 1.1 image-to-video by Alibaba (Tongyi) — generates video from a FIRST FRAME image + prompt. 720P/1080P, up to ~10s. DashScope async.

HappyHorse 1.0 (Video Editing)

HappyHorse 1.0 video-editing by Alibaba (Tongyi) — edits an existing VIDEO based on instruction (prompt) + up to 5 reference images. 720P/1080P. DashScope async (1-5 minutes). ⚠️ Priced by billable video SECONDS = SOURCE length + OUTPUT length (output keeps the same length as the source video); 720P 42cr/second · 1080P 72cr/second.

HappyHorse 1.1 (Reference→Video)

HappyHorse 1.1 reference-to-video by Alibaba (Tongyi) — generates video from 1-4 REFERENCE IMAGES + prompt (preserves character/style). 720P/1080P, up to ~10s. DashScope async.

Claude Sonnet 5.5

Claude Sonnet 5.5 — Anthropic's newest Sonnet model (late September 2026) and the default Sonnet on the Anthropic API. 1M-token context, balancing speed and quality for coding, writing and chat.

Claude Opus 5.5

Claude Opus 5.5 is Anthropic's newest Opus model (released September 22, 2026) and the one Anthropic recommends as the default for most work. According to Anthropic, it comes close to Claude Fable 5.1 on most tasks while producing output faster and using fewer tokens per task than Claude Opus 5 — strong at agentic coding, complex reasoning and professional knowledge work.

Gemini 3.8 Flash

1M free tokens/mo

Gemini 3.8 Flash (Google Vertex) - Flash MOI NHAT dong Gemini 3.x (GA 2/9/2026), thay 3.7 lam ngua tho agentic chinh. Google mo ta la "most intelligent workhorse model yet": tien bo ro o software engineering, tac vu agentic va suy luan nhieu buoc trong linh vuc chuyen sau. LUU Y: 3.8 tieu NHIEU token hon 3.7 de dat hieu nang (nhat la o effort cao) ma token reasoning tinh vao gia OUTPUT - don gia y het 3.7 nhung hoa don co the cao hon; uu tien chi phi/toc do thi dung 3.7. Context 1 trieu token (1.048.576 vao / 65.536 ra). INPUT TOI DA: anh <=3.000/prompt (<=7MB/file inline, <=30MB tu Cloud Storage; png/jpeg/webp/heic/heif) + video <=10/prompt (~45 phut neu CO audio, ~1 gio neu KHONG audio) + audio CHI 1 file/prompt (~8,4 gio hoac toi 1 trieu token) + tai lieu <=3.000 file, <=3.000 trang/file (<=50MB pdf / <=7MB txt). Thinking level LOW/MEDIUM (mac dinh)/HIGH - MINIMAL KHONG duoc ho tro. Grounding Google Search/Maps, code execution, function calling, structured output, URL context, computer use (preview), agentic video understanding (preview). TEXT-only output.

Grok 4.6

1M free tokens/mo

Grok 4.6 — xAI's latest frontier model, with a 500K-token context window and adjustable reasoning effort (low to xhigh). Suited to coding, multi-step reasoning and tool use.

Grok 4.5

1M free tokens/mo

Grok 4.5 — xAI's previous-generation frontier model, with a 500K-token context window and adjustable reasoning effort (low to high). Suited to coding and multi-step reasoning. For the newest model, use Grok 4.6.

Gemini 3.7 Flash

1M free tokens/mo

Gemini 3.7 Flash (Google Vertex) - The LATEST Flash in the Gemini 3.x line (GA 8/13/2026), agentic capability on par with Pro at Flash price/speed, a major leap in code generation + terminal execution. Context 1 million tokens (1,048,576 input / 65,536 output). MULTIMODAL input: text + images (<=3000, 7MB inline/30MB GCS) + audio (<=1 file, ~8.4h) + video (<=10, ~45min with audio/~1h) + pdf/txt documents (<=3000 files). Thinking level LOW/MEDIUM (default)/HIGH - note MINIMAL is NOT supported. Grounding with Google Search/Maps, code execution, function calling, structured output, URL context, computer use (preview). TEXT-only output.

DeepSeek V4.1 Flash

DeepSeek V4.1 Flash — the fast, low-cost DeepSeek V4.1 for high-volume chat and coding.

Kimi K3

Moonshot AI's Kimi K3 — the new flagship, strong at coding, agents and long context.

GLM-5.3

Zhipu AI's GLM-5.3 — an open model strong at coding and agentic tasks, with a thinking mode.

Qwen3.8 Flash

Fast, low-cost Qwen3.8 with a 1M-token context — great for agents, coding and long documents.

Qwen3.8 Max

Qwen's newest flagship (2.4T-parameter MoE) with a 1M-token context — strong at coding, reasoning and professional work.

Claude Fable 5.1

Claude Fable 5.1 (Anthropic, 01/09/2026) is the most powerful model currently available, ranking above Opus 5 — specialized in multi-step reasoning and long-session agentic tasks. Context 1 million tokens, knowledge cutoff 06/2026, adaptive thinking always on. In exchange, it is slower and more expensive than Opus 5; since 22/09/2026 Anthropic recommends Opus 5.5 as the default for most work, and Haiku for short tasks or classification. Note on naming: "Fable" does not mean the model specializes in creative writing — its writing style is denser and less formatted than Opus 5.

Claude Opus 5

200K free tokens/mo

Claude Opus 5 is Anthropic's previous-generation flagship reasoning model — a major step up in complex reasoning, agentic coding, and long-horizon autonomous task execution over the Opus 4.x line, with stronger instruction-following and professional knowledge work. For most new work Anthropic now recommends Claude Opus 5.5.

GPT-6 Astra — OpenAI's strongest model (released 3/9/2026), surpassing GPT-5.6 Sol: multi-step reasoning, computer use, better judgment. Suited for long, end-to-end work (code + app building + research). Context 1.05M tokens. Supports vision (image reading), function calling, and reasoning effort low→ultra — ultra specifically runs multiple subagents in parallel for tasks that can be broken down.

Claude Opus 4.8

Claude Opus 4.8 delivers a major upgrade in complex reasoning, agentic coding, and professional knowledge work. With improved honesty, stronger long-horizon task execution, and a 1-million-token context window on supported platforms, it is built as Anthropic’s most capable generally available model for high-autonomy workflows and demanding enterprise use cases.

Claude Sonnet 5

500K free tokens/mo

Claude Sonnet 5 — Anthropic's previous Sonnet generation: balances speed/quality, strong coding + content, and supports deep reasoning (thinking) and web search. The newest Sonnet is now Claude Sonnet 5.5.

Gemini 3.6 Flash

1M free tokens/mo

Gemini 3.6 Flash (Google Vertex) - near-Pro intelligence at Flash price/speed. Context of 1 million tokens (1,048,576 input / 65,535 output). MULTIMODAL input: text + image (<=3000) + audio (~8.4h) + video (<=10, ~1h) + pdf/txt documents. Thinking (deep reasoning - trace visible via streaming when thinking level is enabled), Grounding via Google Search/Maps/Partners, function calling, structured output. TEXT-only output.

Claude Fable 5

Claude Fable 5 (Anthropic, released 09/06/2026) is the PREVIOUS generation of the Fable line — now classified as legacy by Anthropic, with Fable 5.1 as the current version. Same positioning: multi-step reasoning and long-session agentic tasks, stronger than Opus 5 in that category. Context 1 million tokens, max output 128K, adaptive thinking always on. It differs from Fable 5.1 mainly in knowledge freshness: cutoff 01/2026 (Fable 5.1 is 06/2026). Same price as 5.1, so unless there's a specific reason to pin the older version, Fable 5.1 should be used.

GPT-5.6 Sol

GPT-5.6 Sol — OpenAI's newest flagship coding model (frontier, GA 7/2026): agentic coding, research, multi-step task execution. Supports vision (image reading) + reasoning effort low→ultra.

GPT-5.6 Terra

GPT-5.6 Terra — a balanced variant of the GPT-5.6 series. Vision + reasoning effort low→ultra.

GPT-5.6 Luna

GPT-5.6 Luna — a light/fast variant of the GPT-5.6 series. Vision + reasoning effort low→ultra.

Gemini 3.5 Flash

1M free tokens/mo

Gemini 3.5 Flash (Google Vertex) - near-Pro intelligence at Flash price/speed. Context 1 million tokens (1,048,576 input / 65,535 output). MULTIMODAL input: text + images (<=3000) + audio (~8.4h) + video (<=10, ~1h) + pdf/txt documents. Thinking (deep reasoning - shows trace via streaming when thinking level is enabled), Grounding with Google Search/Maps/Partners, function calling, structured output. TEXT-only output.

Gemini 3.1 Flash-Lite

1M free tokens/mo

Gemini 3.1 Flash-Lite (Google Vertex) - the cheapest and fastest in the Gemini 3.x family. Context of 1 million tokens (1,048,576 input / 65,535 output). MULTIMODAL input: text + image (<=3000) + audio + video + pdf/txt documents. Thinking (with visible trace), Grounding via Google Search/Maps/Partners, function calling, structured output. TEXT-only output.

Gemini 3.1 Pro

1M free tokens/mo

Gemini 3.1 Pro (Google Vertex) - reasoning/chat LLM with 1 million token context (1,048,576 input / 65,536 output). MULTIMODAL input: text + images + audio + video + documents. Thinking + Grounding with Google Search/Maps/Partners + function calling. TEXT-only output.

Claude Opus 4.7

Claude Opus 4.7 delivers a quantum leap in logical reasoning and creative synthesis. With its next-gen architecture and massive 2-million-token context, it serves as the ultimate high-performance partner for those pushing the boundaries of innovation and discovery.

Multimodal Embedding 001

Multimodal Embedding 001 (Google Vertex) — embeds TEXT and/or IMAGE into the SAME 1408-dimensional vector space (search images by text, image-text matching). Input: text and/or image.

Gemini Embedding 001

Gemini Embedding 001 (Google Vertex) - the LATEST text embedding model, 3072-dimensional vectors. Billed per token (= Google pricing). Text input; supports batch processing of multiple texts.

Midjourney Image-to-Image

Midjourney Image-to-Image (i2i) — uses ≥1 reference image(s) (image_urls) to guide style/structure + a descriptive prompt → 4 PNG images. Each URL = 1 image prompt (MJ auto re-hosts, recommended ≤5). Async ~30-60s. Advanced: --sref/--cref in prompt.

Seedream 5.0 Pro (Image to Image)

Image-to-image with Seedream 5.0 Pro. Send 1–10 reference images + a prompt describing the desired change → returns 1 transformed image, no watermark. Different from the text-to-image version: t2i generates 4 images per request, i2i only 1 image.

GPT Image 2(img2img)

GPT Image 2 is OpenAI’s next-gen image model built for stronger photorealism, cleaner image editing, sharper text rendering, and more polished product photography. Designed for more advanced visual workflows, it pushes image generation beyond basic text-to-image output and into higher-quality creative, commercial, and design-ready use cases.

Qwen-Image-Edit Plus

Qwen-Image-Edit Plus by Alibaba — edit images by instructions, balancing quality/cost (Plus tier, cheaper than Max). Accepts 1–3 reference images + instruction.

Qwen-Image-Edit Max

Qwen-Image-Edit Max by Alibaba — edit images by instructions, the highest quality in the edit lineup (Max tier). Accepts 1–3 reference images + instruction.

Wan2.2 I2V Flash (Ảnh→Video 480P)

Wan2.2 I2V Flash — a silent 5-second clip from one image; 480P costs just 25 credits. The cheapest lane for testing ideas.

Wan2.6 I2V Flash (Ảnh→Video, có tiếng)

Wan2.6 I2V Flash — the budget Wan image-to-video: pick silent or with audio, 2–15 seconds. Silent 720P is just 8 credits per second.

Wan2.2 KF2V Flash (Frame đầu → frame cuối)

Wan2.2 KF2V Flash — give a first and last image and the model builds the transition between them. Output is always 5 seconds.

Wan2.2 Animate Mix (Thay nhân vật trong video)

Wan2.2 Animate Mix — swaps the character in an existing video for the one in your image, keeping the original motion and scene. Billed by output length.

Wan2.2 Animate Move (Nhân vật nhại động tác)

Wan2.2 Animate Move — give one character image plus a driving video; the character mimics the motion. Billed by the output video length.

Vidu Q2 Pro Fast (Ảnh→Video)

Vidu Q2 Pro Fast — the cheapest, fastest video lane available: one image to a 720P/1080P clip at 5 credits per second.

Vidu Q3 Drama (Phim / kịch)

Vidu Q3 Drama — the film/drama variant: acting, character emotion and cinematic cuts; 1080P costs the same as 720P.

Vidu Q3 Ad (Video quảng cáo)

Vidu Q3 Ad — the product-advertising variant: keeps the exact shape, label and material of the product in your reference images.

Vidu Q3 Mix (Tham chiếu→Video)

Vidu Q3 Mix — feed 1–7 reference images (characters, props, scenes) to build a video that keeps their identity. Up to 16 seconds.

HappyHorse 1.1 (Image→Video)

HappyHorse 1.1 image-to-video by Alibaba (Tongyi) — generates video from a FIRST FRAME image + prompt. 720P/1080P, up to ~10s. DashScope async.

HappyHorse 1.1 (Reference→Video)

HappyHorse 1.1 reference-to-video by Alibaba (Tongyi) — generates video from 1-4 REFERENCE IMAGES + prompt (preserves character/style). 720P/1080P, up to ~10s. DashScope async.

ElevenLabs Scribe v2

Chuyển giọng nói thành chữ với độ chính xác cao, chấm câu tốt, nhận 99 ngôn ngữ trong đó có tiếng Việt. Tính tiền theo THỜI LƯỢNG audio nguồn.

Fun-ASR (Speech to Text)

Fun-ASR (Alibaba DashScope) - converts speech to text, multilingual (Chinese/English/Vietnamese...). Accepts public audio URLs (wav/mp3), returns transcript. Asynchronous via transcription API.

ElevenLabs Forced Alignment

Canh mốc thời gian cho từng TỪ và từng KÝ TỰ giữa audio và lời thoại đã biết — dùng làm phụ đề khớp môi, karaoke, highlight theo câu. Tính tiền theo thời lượng audio nguồn.

Seedream 5.0 Pro (API)

Seedream 5.0 Pro on the official BytePlus API — 1K or 2K output, up to 4 input images.

Vidu Image (Tham chiếu→Ảnh)

Vidu Image — generates images from multiple reference images while preserving characters and props. Up to 4K.

CinAI Nano Pro

5 free runs/mo

CinAI Nano Pro — reasoning image model: accurate text rendering, real-world knowledge, image creation and editing with up to 14 reference images, web search grounding and system instructions. 1K/2K/4K resolution, 10 aspect ratios.

Nano Banana Pro

20 free runs/mo

Nano Banana Pro (Google Labs Flow) — Google's premium image model with accurate text rendering and real-world knowledge. Text-to-image or image-to-image from reference images. 5 aspect ratios, 1 image per run. Async.

CinAI image-elements

Generate an image and instantly split every object into transparent PNG layers in one run. Prompt only = text-to-image (4 images); add input images for image-to-image. Returns the full image plus each cut-out layer and a .zip bundle. Note: only one of the 4 images gets layered (see sourceUrl); occluded objects are cut out only where visible.

Grok Image

30 free runs/mo

Create/edit images via grok.com/imagine. Text-to-image (prompt only → 4 images) or image-to-image (with 1–3 input images → 1 edited image). Aspect Ratio applies to text-to-image; when an input image is provided, the output KEEPS the original image's aspect ratio (aspect setting is ignored). Requires a SuperGrok plan.

Qwen Image 3.0 Pro

Qwen-Image 3.0 Pro — the premium Qwen-Image 3.0 with finer detail and composition, up to 2K (2048×2048).

Qwen Image 3.0

Qwen-Image 3.0 — the newest generation with accurate in-image text rendering, photoreal images and posters. Supports reference images.

Midjourney Image

Midjourney Image (t2i) — generates 4 high-quality PNG images from a prompt via a browser gateway (Midjourney Standard subscription). Async ~30-60s. Supports aspect_ratio, version, stylize, weird, chaos, raw, speed, negative.

Seedream 5.0 Pro

Text-to-image & image-to-image with Seedream 5.0 Pro (ByteDance). t2i: prompt only → 4 images 2431×1365. i2i: add up to 10 reference images → 1 image. Images have NO watermark, hosted permanently on media.halcion.vn.

Nano Banana 2

50 free runs/mo

Nano Banana 2 (Google Labs Flow) — balanced speed and quality, lighter than Nano Banana Pro. Text-to-image or image-to-image from reference images. 5 aspect ratios, 1 image per run. Async.

CinAI Nano 2

10 free runs/mo

CinAI Nano 2 — sharper 2K imagery, intelligent 4K upscaling, improved text rendering and stronger character consistency. A major leap in visual quality for creative work and API-driven workflows.

CinAI Nano Lite

30 free runs/mo

CinAI Nano Lite — the lightest and fastest model in the CinAI Nano line (~4s per image). Generates images from a prompt plus up to 14 reference images (image-to-image). 1K resolution (~1MP), png/jpeg/webp output.

Nano Banana 2 Lite

100 free runs/mo

Nano Banana 2 Lite (Google Labs Flow) — the lightest and fastest in the Nano Banana line. Text-to-image or image-to-image from reference images. 5 aspect ratios, 1 image per run. Async.

CinAI Nano

20 free runs/mo

CinAI Nano — low-cost image generation and editing model, strong at editing and blending images. Up to 3 reference images, 10 aspect ratios, fixed ~1024px resolution. Scheduled for retirement on 2026-10-02.

GPT Image 2

GPT Image 2 is OpenAI’s next-gen image model built for stronger photorealism, cleaner image editing, sharper text rendering, and more polished product photography. Designed for more advanced visual workflows, it pushes image generation beyond basic text-to-image output and into higher-quality creative, commercial, and design-ready use cases.

Z-Image Turbo

Z-Image Turbo by Alibaba (Tongyi) — an ultra-lightweight, high-speed, low-cost text-to-image model; renders Chinese/English text, with flexible resolution of 512–2048px. Fixed at 1 image per generation. Prompt up to 800 characters.

ElevenLabs Music

Sáng tác nhạc từ mô tả bằng chữ, dài từ 3 giây tới 10 phút. Tính tiền theo GIÂY nhạc tạo ra.

Lyria 2

Lyria 2 (Google Vertex) — generates instrumental MUSIC from English prompts. ~30s clip, WAV. Accepts English prompts only.

ElevenLabs Sound Effects

Sinh tiếng động từ mô tả bằng chữ: mưa rơi, cửa mở, tiếng bước chân, nền không gian… Tính tiền theo GIÂY audio tạo ra.

Lyria 3 Clip

Lyria 3 Clip (Google Vertex, Preview) — SHORT ~30s clip variant of Lyria 3. text->music & image->music, vocal/instrumental, BPM/Intensity/Lyrics WITHIN the prompt (NO Duration parameter). Output MP3.

Lyria 3 Pro

Lyria 3 Pro (Google Vertex, Preview) — generates 3rd-generation MUSIC: full song (music+lyrics+vocals), text->music & image->music, up to 184s. Controls BPM/Intensity/Duration/Instrumental/Lyrics/Structure WITHIN the prompt. Output: MP3 + lyrics + description.

Gemini 3.1 Flash TTS

Gemini 3.1 Flash TTS (Google Vertex, Preview) - 3.1-generation TTS, high quality. 30 voices + 78 BCP-47 languages. WAV 24kHz.

ElevenLabs v3

74 ngôn ngữ · biểu cảm nhất

Gemini 2.5 Flash Lite TTS

Gemini 2.5 Flash Lite TTS (Google Vertex, Preview) - the cheapest TTS, high-volume. 30 voices + 78 languages. WAV 24kHz.

ElevenLabs Multilingual v2

29 ngôn ngữ · ổn định, chuẩn ngành

Gemini 2.5 Flash TTS

Gemini 2.5 Flash TTS (Google Vertex) - fast & cheap text-to-speech. 30 prebuilt voices (Achernar...Zubenelgenubi, male/female) + 78 BCP-47 languages. Output WAV 24kHz. Asynchronous via generateContent + speechConfig.

ElevenLabs Flash v2.5

32 ngôn ngữ · độ trễ ~250ms, rẻ một nửa

ElevenLabs Turbo v2.5

Giọng đọc đa ngôn ngữ độ trễ rất thấp, chi phí bằng một nửa Multilingual v2. Hợp với trợ lý thoại, tổng đài và mọi việc cần trả lời gần như tức thì.

ElevenLabs v3 Conversational

Biến thể hội thoại của v3: ngữ điệu tự nhiên như người đang nói chuyện, độ trễ thấp và giá bằng một nửa v3 thường.

Gemini-2.5-pro-tts

text-to-speech

Qwen3-TTS Instruct Flash

Qwen3-TTS Instruct Flash by Alibaba — text-to-speech with style CONTROL via instructions (speed, tone, emotion). 19 voices, multilingual. Returns WAV.

Seedance 1.0 Pro Fast

Seedance 1.0 Pro Fast — the fast, cheap Seedance 1.0 Pro; 720p costs just 7 credits per second.

Seedance 1.0 Pro

Seedance 1.0 Pro — ByteDance's steady video generation, 480p to 1080p, up to 12 seconds.

Seedance 2.5 (API)

Seedance 2.5 on the official BytePlus API — 480p to 1080p, 4–15 seconds, with reference image/video/audio input.

CinAI Gemini Omni

1 free runs/mo

CinAI Gemini Omni — text-to-video with audio and a thinking step. Supports image-to-video (start image), 16:9 / 9:16 aspect ratios, up to 10 seconds.

Omni 1.1 Flash

10 free runs/mo

Omni 1.1 Flash (Google Labs Flow) — video with audio from text, or referencing 1–4 character/object images and one video clip. 16:9 / 9:16, 4/6/8/10 second duration. Async.

Grok Video 1.5

5 free runs/mo

Generate video via grok.com/imagine (browser, requires SuperGrok). Text-to-video (prompt only) or image-to-video (with 1 image). Aspect Ratio applies to text-to-video; with image input (i2v), the original image's aspect ratio is kept.

Wan3.0 Video Prime

Wan3.0 Video Prime — the premium Wan3.0 with more lifelike visuals and consistent characters, videos up to 30 seconds with sound.

Wan3.0 Video

Wan3.0 Video — Alibaba's newest all-in-one video model: text-to-video, first/last-frame and reference-image video up to 30 seconds, with sound.

Seedance 2.5

Text-to-video & image-to-video via Dreamina Seedance 2.5 (ByteDance) — the LATEST version, highest quality (Dola: "For best video quality"). Pass a prompt (t2v) or add reference images (i2v, up to 10 images). Duration 5/10s, 6 aspect ratios. Async: 5s ~1-3 minutes, 10s ~5 minutes.

Seedance 2.0

Text-to-video & image-to-video via Dreamina Seedance 2.0 (ByteDance) — browser automation. Pass a prompt (t2v) or add a reference image (i2v). Duration 5/10s, multiple aspect ratios. Async 1-3 minutes.

CinAI Veo 3.1

CinAI Veo 3.1 — CinAI's highest-quality video model. Text-to-video and image-to-video (image ≤20MB, first + last frame). 4/6/8 second duration, 720p/1080p, 16:9/9:16, 24 FPS, with audio. Async (~60-90s).

CinAI Veo 3.1 Fast

1 free runs/mo

CinAI Veo 3.1 Fast — fast, cost-effective video generation. Text-to-video and image-to-video (image ≤20MB, first + last frame). 4/6/8 second duration, 720p/1080p, 16:9/9:16, 24 FPS, with audio. Async (~40-70s).

Dreamina-Seedance-2.0-mini

Multimodality-to-video generation model by ByteDance/Doubao: takes text + image + video + audio references → video (may include audio). 480P/720P, 4-15s, 24fps.

Dreamina-Seedance-2.0-fast

Seedance 2.0 Fast by ByteDance/Doubao — faster multimodal VIDEO generation: text + images (1-9) + video (0-3) + audio (0-3) references → 1 video (may include audio). 480P/720P, 4-15s, 24fps; modes include Text→Video, Image→Video, First/Last Frame, Multimodal Reference.

Dreamina-Seedance-2.0

Seedance 2.0 (full version) by ByteDance/Doubao — generates multimodal VIDEO: text + image (1-9) + video (0-3) + audio (0-3) references → 1 video (which may include audio). Supports 480P/720P/1080P/4K, 4-15s, 24fps; modes include Text→Video, Image→Video, First/Last Frame, and Multimodal Reference.

Veo 3.1 Lite

10 free runs/mo

Veo 3.1 Lite (Google Labs Flow) — lightweight Veo 3.1 text-to-video with audio. 16:9 / 9:16, 4/6/8 second duration. Async.

CinAI Veo 3.1 Lite

2 free runs/mo

CinAI Veo 3.1 Lite — the most affordable model in the CinAI Veo line (Preview). Text-to-video and image-to-video (first frame). 4/6/8 second duration, 720p/1080p, 16:9/9:16, 24 FPS, with audio. Async (~40-70s).

HappyHorse 1.1 (Text→Video)

HappyHorse 1.1 text-to-video by Alibaba (Tongyi) — generates video from a PROMPT. 720P/1080P, ratios 16:9/9:16/1:1/4:3/3:4, up to ~10s. DashScope async.

ElevenLabs Audio Isolation

Tách giọng nói ra khỏi tiếng ồn và nhạc nền, trả về bản thu sạch. Audio nguồn phải dài ít nhất 4,6 giây. Tính tiền theo thời lượng.

ElevenLabs Voice Changer

Giữ nguyên ngữ điệu, nhịp và cảm xúc của bản thu gốc nhưng đổi sang một giọng khác. Tính tiền theo thời lượng audio nguồn.

ElevenLabs Dubbing

Lồng tiếng bản thu sang ngôn ngữ khác mà GIỮ NGUYÊN chất giọng người nói gốc, kèm phụ đề .srt đã dịch. Bản trên gateway KHÔNG có watermark. Nguồn tối đa 10 phút. Tính tiền theo thời lượng.

HappyHorse 1.0 (Video Editing)

HappyHorse 1.0 video-editing by Alibaba (Tongyi) — edits an existing VIDEO based on instruction (prompt) + up to 5 reference images. 720P/1080P. DashScope async (1-5 minutes). ⚠️ Priced by billable video SECONDS = SOURCE length + OUTPUT length (output keeps the same length as the source video); 720P 42cr/second · 1080P 72cr/second.