The ultimate AI short film tool is here!
Friends, would you believe it? I uploaded the original text of a novel and directly generated this comic drama:
And this one-minute MV: I uploaded character photos and music, and the AI automatically designed the scenes and visuals for me, and even completed the editing.
Recently, ByteDance's Seedance 2.5 officially launched, and this can be called an epic-level upgrade. A single generation supports up to 30 images, 10 videos, and 10 audio clips as references, and the maximum single-generation duration has also increased from 15 seconds to 30 seconds. Instruction understanding and video editing capabilities have both improved.
Judging by model capability alone, Seedance 2.5 is already in the current top tier.
But strong performance does not equal a great experience in use。
Generating a 30-second short film in one go looks great,but even if just one frame does not match what I have in mind, I have to spend nearly 40 yuan to regenerate the entire video. Honestly, using it hurts my wallet... I do not even dare to casually try my luck with a 30-second video.
Today our old friend LibTV officially integrates Seedance 2.5. During the limited-time event, Seedance 2.5 is actually only 0.4 yuan/second, and a single generation can directly make a super-long 5-minute video. If there is a problem with 2 seconds in the middle, you can regenerate just those 2 seconds, and the usage cost drops immediately. You can completely let loose and use it freely!
The key point is,LibTV is an official Volcano partner, and all generation tasks can transparently query the taskid, making the generation process traceable.
In addition, LibTV has recently launched several especially useful new features! Let us test and experience them together.
01. Hands-on test of LibTV Seedance 2.5
Smart referencing
Seedance 2.5 supports up to 30 images, 10 videos, and 10 audio clips as references in a single generation, but with so much material uploaded, adding references one by one in the prompt is especially troublesome.
LibTV specifically created a smart referencing feature for this. We connect the materials on the canvas, and when entering the prompt, LibTV will automatically match the corresponding references on the canvas.
We only need to click or press TAB to confirm.
Seedance 2.5 supports up to 50 references. A relatively complex project may require dozens or even hundreds of material matches. Manually adding them one by one would take at least several minutes, while LibTV's smart referencing can get it done in one second.
For example, if I want to make a street photography Vlog, we paste the prompt or script directly in, and LibTV will automatically match the materials on the canvas and display the reference locations. We can press Shift + TAB to directly insert all references, super convenient!
Seedance 2.5 has a very good grasp of characters and lighting. The characters' skin and action poses are very realistic, and the speaking tone also fits real-world scenes very well.
Let us send the same prompt for the first 15 seconds to Seedance 2.0 and try:
Seedance 2.5 has a very good grasp of characters and lighting. The characters' skin and action poses are very realistic, and the speaking tone also fits real-world scenes very well.
Let us send the same prompt for the first 15 seconds to Seedance 2.0 and try:
Seedance 2.0 Although there are no major bugs, the characters look obviously AI-generated, and when the character speaks, they inexplicably look to the right, with unnatural eyes and movements.
Seedance 2.5 Skills
To reduce the gacha odds, LibTV has built a professional Skill library around Seedance 2.5. For professional film and TV, commercial ads, music MVs… we can directly invoke professional workflows without having to obsess over prompts ourselves.
For example, we used the POP MV Skill to make an MV for Kill This Love:
Prompt: Help me make a 1min video for this song. The female lead is this person, and use this song for the music. When generating the video later, please be sure to feed the song's audio in segments to the video as audio reference for generation.
I uploaded the music and the character's photo at the same time. LibTV will follow the Skill's kickoff process and first precisely understand my requirements through 9 detailed questions:
LibTV automatically splits the soundtrack on the canvas, generates character looks and scene images, completes storyboard planning, and then edits it into a film.
The final generated MV looks like this:
The scene, lighting, and shot design are all very mature, and the switches in shots and lighting basically land on the rhythm of the music. The character movements, expressions, and stage effects all have an obvious sense of design, and the whole MV looks very professional.
Frame-by-frame breakdown
Creating something is inseparable from borrowing from good films. Upload a video in LibTV and useFrame-by-frame breakdownto automatically break it into a whole set of asset groups, andit consumes no compute at all！
LibTV breaks down video materials from three angles: storyboard, motion, and music.
There is not only breakdown of storyboard images, but also breakdown of video clips. The image breakdown, camera positions, and camera movement methods are all broken down in especially detailed and accurate ways;
We can quickly understand the basic narrative thread, character relationships, and visual style of a reference film, and easily generate a film with a similar tone.
We open the canvas Agent on the right and enter:
Refer to the overall tone of the film and generate a film in a similar style. Replace the scene with image 1, replace the male lead with image 2, and replace the female lead with image 3.
The shot language of the reference film is basically preserved, and the shot size transitions, camera position arrangements, and camera movement rhythm are all very natural and very cinematic.
Ultra-long video
Seedance 2.5 in LibTV breaks through the ceiling and can directly output a 5-minute ultra-long video, making it super suitable for AI comic dramas and short dramas.
Prompt: Based on the novel I provide, generate a 5-minute ancient-style comic drama, preserving the core plot, character relationships, important dialogue, and satisfying moments of the original work; the art style should adopt a next-generation Chinese-style game CG cutscene animation style, with refined characters, delicate clothing patterns, and gorgeous and meticulously researched ancient scenes, using cinematic composition, volumetric lighting, fine materials, and smooth camera movement, and ensure that the character appearances and scene designs remain consistent throughout the entire film.
LibTV will directly help us generate the complete video.
We clickView generation processat the upper right of the video, and we can expand to view all nodes, connection relationships, and prompts. Every reference image and shot can be modified individually.
Clip reshoot
If we are not quite satisfied with a certain generated image, we do not need to reroll the entire clip like before.
We select the video and clickClip reshoot。
Select the clip that needs to be modified in the timeline, and enter the editing idea below:
LibTV will make precise modifications to the timeline clip we marked, and the other parts of the image will be exactly the same as the original film.
For this 26-second video, if rerolling, it would take 1196 credits, but using Clip reshoot for 7 seconds only uses 374 credits.You only spend credits for the length of footage you change, so you don't waste credits for nothing.。
Let's take a look at the final finished result together:
The reference short film's shot language and visual tone are basically preserved; the shot size transitions, camera placement, and camera movement rhythm are all very natural and have a strong cinematic feel.
02. Some Sharing
Seedance 2.5 has more refined visuals; characters' expressions, movements, and lighting are closer to real filming, reference materials are restored more accurately, and characters and scenes remain more stable across continuous shots.
But everyone can see how expensive Seedance 2.5 is; almost 1 second of video costs more than 1 yuan.
Generating 30 seconds in a single pass looks more satisfying, but every time I click generate, I secretly worry whether there will be flaws and whether some shots will waste dozens of yuan.
The longer the video, the higher the chance of errors, and the higher the cost of revisions. Many ideas are discouraged by cost before they even start being tried.
LibTV has built a series of exclusive features around the AI video creation workflow, such assmart referencing, Seedance 2.5 professional Skills, free frame-by-frame shot analysis, single-pass generation of ultra-long 5-minute videos, and clip reshootsturning AI video into a project that can be continuously revised.
Wherever you're not satisfied, reshoot just that part; shots you're already happy with don't need to be paid for again.
In addition, during the event, Seedance 2.5 is available for a limited time at as low as 0.4 yuan/second, bringing down both generation and revision costs.
We can finally work freely, try several approaches, and slowly polish the work until we're satisfied.
Original link:After trying LibTV's new features, I finally dare to freely use Seedance 2.5