Novel synthesized video, working in four parts:
- Novel Generation Images
- Novel Generation Audio
- Novel Synthesis Video
- Secondary refined video
We have learned from our previous teachingDemystifying AI Comic Book Narration Video from Principle to Practice (Making Novel Scripts)Beginning, making a novel script At the end, we get a stitched EXCEL, please download it or generate it yourself according to the tutorial.
Let's look at the finished product first
I will follow the procedure of the software I use to demonstrate the process. (Tools are not important and again, no paid software is recommended)
Novel Generation Images
1. Start creating
2. Before creating a novel, test and debug the mapping environment first
Make sure the following debugging is OK, then click "Create Project".
3. Create the first step of the project, select Import Script File
4. Select the previously saved EXCEL (can not find turn over the previous article)
5. Check the import file, because all the text has been processed in EXCEL, this step does not need to adjust anything!
6. Pay attention to the "Setting Character" in the upper right corner, we can still make secondary corrections and attach other LoRA styles!
Since I've already debugged the style locally in SD before importing the file, I won't modify it here.
After saving, go to the next step and let's move on to generating images.
7. Enter the picture generation interface, do not click on "start generating pictures", let us check the configuration again!
8. Some of my configurations
I didn't go into detail about my configuration in step 2, so I'll simply list the main items here:
- Model: Ouka - National Style
- vae: vae-ft-mse-840000-ema-pruned
- Positive presets: (best quality, 4k, 8k, highres, masterpiece:1.2),(ultra-detailed),masterpiece,(masterpiece, top quality, best quality),
- Negative: not showing up, words easily violated, randomly copied
- Amplification algorithm: R-ESRGAN 4x+
- Image size: long (Long map will have all kinds of a problem, many models suggest generating square map, 512 * 512 can be)
In the above figure, any lens, the list of the right side of the "Adjust Parameters" button, click on it to enter the detailed parameters of the current lens configuration page.
Test generating images for one or two lenses, feel fine, and start generating images in bulk.
9. Batch generate images
Up to the current position, 37 charts, 3060 cards, took about 3 minutes.
Drawing the highlights: installing the Tensorrt extension and generating the ONNX model can theoretically be compressed into a 2-minute finish
10. One-click lossless zoom
Because the default generation is a small size picture (generation speed is fast and convenient to adjust it), if these pictures are basically in line with the requirements, do not need to repair (here I do not repair), click on the non-destructive enlargement.
At this point we have got a set of pictures, you can start automatically synthesize the video, I demonstrated the tools to follow up the synthesis of voice, video process in the software have, but I demonstrated the use of three-way audio generation tools here, read aloud better.
These pictures go here first... First, we'll fix the reading of the novel.
Novel Generation Audio
Here I don't use the common Microsoft programs, or Tease. All things considered it's better to go with the free but more realistic vocals sound clippers The
We saved the following novel and went to the sound clipping to start the process...
I'm dressed as a fairy.
Not the fairy of mortal legend, but the real kind that flies around in the sky. The Emperor of Heaven is my father and the Queen of Heaven is my mother, so I was born a fairy, with white skin and long legs, and full of fairy spirit.
The days of a fairy are simply too comfortable. Getting high, reading books, and playing with the little fairies in their own fairy palace all day long is awesome.
Unfortunately, the good times didn't last long before something happened.
That afternoon, my palace's little fairy servant, Li Li, came running in with great haste, "White Lotus Supreme Immortal! White Lotus!"
"Hear hear hear, stop yelling." I'm still not quite used to the name, even though it's been months since I crossed over.
"Immortal Lingfeng has returned from his descent into the mortal world to cross the tribulation, let's go and visit."
"Who is Immortal Lingfeng?"
"Supreme Immortal are you sleepy, that's your fiancé!"
"Huh??"
It was then that I remembered that I seem to have a fiancé. Why does the name Ling Feng sound so familiar, it's exactly the same as the hero of the novel I read before I crossed over?
After some inquiries, I finally figured out the situation: I, White Lotus, am the female counterpart in this abusive text.
The main character, Ling Feng, is an upper immortal in the immortal world, reaching the position of upper immortal in 20,000 years, a rare talent in the immortal world. Although I am also an Upper Immortal, unlike him who relied on his own skills to cultivate, I am an Immortal Second Generation, born as an Upper Immortal.
My Heavenly Emperor dad saw Ling Feng's talent, and at my urging he hit it off with Ling Feng's master, Ancestor Bodhi, and set up a marriage between the two of us.
But Lingfeng was not at all satisfied with the marriage, and also had a big fight with the Bodhi ancestor, and finally parted on bad terms, turning around and going down to the mortal world to experience the calamity.
I counted the time, this time, the robbery, Ling Feng reincarnated into a cultivation sect, and the female lead is his personal disciple. He didn't go back to the immortal world immediately after the calamity, but stayed on earth for a few more decades and waited until the female lead, Qian Yue, succeeded in ascending before they came back together.
The original owner originally had a good feeling for Ling Feng as a disciple as well, but when she found out that Ling Feng's feelings for Qian Yue were more than just master and disciple, she started to go crazy with her death wish, trying to find ways to break up the two of them.
In the end she couldn't help but be jealous and pushed Qian Yue off the Immortal Exorcism Platform, almost sending people's souls flying. After the truth came out she was expelled from the Immortal Realm and was also killed by Ling Feng himself and her soul was dispersed.
The thought of it gave me a terrible headache. I don't want to meet someone who's going to beat the hell out of me later.
But in order not to collapse my persona, I was still dragged by Lili to the Lingfeng Palace for the banquet.
Lingfeng Palace was built next to Peach Blossom Valley, and it was a good time to enjoy the flowers. I swiped my face to enter the palace gate and slipped out of the banquet in the twinkling of an eye. Passing by the liquor place, I smoothed down an altar of Peach Blossom Brew and ran to the Peach Blossom Valley to drink and enjoy the flowers.
Peach Blossom Brewing really lives up to its name, it's very good to drink, but it has a strong aftertaste. In a while I got sleepy, and found a tree to lie down and squint for a while.
In a daze, I heard a movement, I thought it was Ri Ri looking for me, so I poked my head out of the tree and prepared to pamper myself. But it wasn't Ri Ri who came, it was the heroine Chiyuki.
Qian Yue obviously didn't expect this someone and froze, looking at me with tearstains still on her face. It was then that I remembered the plot written in the novel where the heroine was ridiculed for her looks when she first came to the immortal world and ran out in tears.
Lol, it's childish to make fun of looks when you're thousands and thousands of years old immortal.
As a qualified White Lotus, ah no, White Lotus Upper Immortal, how could I not care when I saw a cute girl being bullied and crying? I snapped my fingers and summoned the flower demons, making them dance around Chiyuki. Qian Yue had never seen this battle before, her face instantly turned red.
She looked at me cautiously, and I smiled at her, hearing RiRi calling out to me in the distance, and beckoned a cloud to go home with RiRi.
Within a few days, my Heavenly Stepmother came to visit me.
"Son, have you seen Ling Feng?" The Heavenly Empress smiled extraordinarily kindly.
"Ah... No." I slipped out early and didn't see anyone at all.
"This time he came back from the calamity, his cultivation has made great progress again, your father said that he has a great future, you and him..."
Gotta be, it's a rush to get married.
1.Upload your work
2. Editorial readers
It automatically segments and recognizes the reader after importing, but it's a real pain to edit that way. And it makes the browser lag.
Here is a trick, because we want to quickly batch production, the novel only needs a person to read aloud the narrator, and the software automatically recognized as a multi-reader, editing is very cumbersome, we can just paste the novel area to enter a sentence "111", and then enter the editing interface, so that you can replace the narrator area in the 111 for the complete! The content of the novel.
The downside is that it doesn't recognize polysyllabic words very well; note that the picture above automatically recognizes polysyllabic words for you.
There is actually another template import solution that works perfectly, explore it on your own, the tool is already given to you...
2. Select the speaker and get the synthesized audio, the process is not shown.
According to the style of novels with female protagonist, reversal and crossing, we choose the middle-aged calm imperial voice, the speed of speech should not be too fast, 1.3 is more appropriate, and the tone of voice is raised.
3. Use silhouette to get audio SRT file
The process is not demonstrated, the subtitles are recognized after uploading the audio, the only thing to note here is: select "Text Match".
By this point you have the audio and the corresponding SRT file
Back to the composite video process...
Novel Synthesis Video
The composite video still relies on software, which matches the timing of the novel's subtitle text reading with the timing of the picture switching, and then composites it together.
What about sound? Don't worry, with subtitles and corresponding picture switching times, sound is synthesized into the picture at any time. Follow the flow of the tool as I demonstrate...
1. Select the draft cutout to generate
2. Choose to upload SRT subtitles
3. Select the effects (screen switch effects), and then click on the draft synthesis
4. Open the clipping, you will see the draft, open it...
5. Import Audio
By this point have got the complete video, the effect is still almost, the last part let us beautify the following video. I am not familiar with editing video, just look at the effect.
Secondary refined video
1. Setting the screen ratio
Usually 4:3 is good, here I purposely do vertical close-ups of real people, so I use 9:16
2. Choose a background fill to flesh out the image
3. Insert some keyframes or special effects to make static pictures look more dynamic
4. Insert marketing cover
Generally not suitable for these types of AI comics, better suited for looping with a pop-up brainwashing video and then playing the novels
6. Add some BGM... It's more brainwashing.
Loop the music in the background. Keep it down.
7. it's not enough... For traffic, we need to add a no-brainer cover.
This method is very suitable for short plays, short articles, do novel serialization can also try this method, you can absorb the volume, drop weight.
Lastly: Packaging some of the resources used throughout the process
Demystifying AI Comics Article Series Table of Contents
- Demystifying AI Comic Interpretation Video from Principle to Practice (Principles)
- Demystifying AI Comic Explainer Videos from Principle to Practice (Sub-Commissioning Channels Chapter)
- Demystifying AI Comic Book Narration Video from Principle to Practice (Acquiring and Rewriting Fiction)
- Demystifying AI Comic Book Narration Video from Principle to Practice (Making Novel Scripts)
- Demystifying AI Manga Narration Video from Principle to Practice (Composite Video)