-
-
Notifications
You must be signed in to change notification settings - Fork 365
FREE Suno Killer I Tested ACE Step 15 XL on My PC
Full tutorial link > https://www.youtube.com/watch?v=2PVmRObV6xI
Newer better full tutorial (17 June 2026) : https://youtu.be/9C_6qNKjgpA
Install and use ACEStep 1.5 XL Premium, a locally runnable, open-source AI music and song generation application for Windows, RunPod, Massed Compute, SimplePod and Linux. In this tutorial I show the default song generation quality, explain the benchmark comparison with Suno 5, walk through the one-click Windows installation, and demonstrate how the app automatically chooses the best model settings for your GPU.
Links:
Download App and the source post: [ https://www.patreon.com/posts/ace-step-1-5-xl-premium-157675060 ]
Windows Requirements tutorial: [ https://youtu.be/DrhUHnYfwC0 ]
You will learn how to download the latest zip, run the install/update BAT, start the interface, download all models, recover interrupted downloads, use GPU tiers, set lyrics, styles, vocals, language, duration, seeds, image-to-video output, MP3/FLAC saves, library history, batch processing, enhance lyrics/style tools, advanced remix, repaint, LoRA, LoRA training, dataset preparation and custom presets.
Main topics include local AI music generation, ACEStep XL 1.5 Turbo, SFT and Base models, one-click Windows setup, GPU tier presets, automatic model downloading, batch song creation, remix, repaint, LoRA support, LoRA training and repeatable saved presets.
Use the chapters to jump to install steps, cloud setup, generation settings, batch tools and advanced app features.
Chapters:
00:00:00 ACEStep 1.5 XL Premium intro and what this local AI music app does
00:00:24 Published benchmark comparison with Suno 5 and quality overview
00:01:02 Default one-click generation demo setup with demo lyrics and style
00:01:35 Remix and repaint support before starting the full tutorial
00:02:14 Generated song demo begins with the first verse
00:02:57 Demo song second verse and chorus continuation
00:03:52 Demo song bridge, final chorus and quality showcase
00:05:10 Results recap, commercial-use quality and tutorial roadmap
00:05:35 Download the latest installer zip from the tutorial post
00:06:00 Read the post and complete the one-time Windows requirements
00:06:22 Supported platforms: RunPod, Massed Compute, SimplePod and Linux
00:06:37 Choose an install folder and avoid spaces or weird path characters
00:06:58 Extract the zip and identify the installer files
00:07:13 Install/update BAT creates the virtual environment automatically
00:07:39 Automatic Turbo model download and supported ACEStep models
00:07:56 Automatic best parameters for each model and GPU tier
00:08:23 Main UI controls: language, male/female vocals and instrumental mode
00:08:36 Duration, song count, random seed and reproducible generation
00:09:15 Image-to-video generation, resolution and aspect ratio handling
00:09:31 Outputs folder, full metadata, MP3 files and FLAC files
00:09:52 Fresh installation completed with verified model downloads
00:10:14 Start the app with the Windows start app BAT file
00:10:27 Download all models and repair interrupted model downloads
00:11:20 Massed Compute installation instructions and command order
00:11:52 RunPod and SimplePod templates plus CUDA 13 driver warning
00:12:39 Linux installation path and first recommended test generation
00:13:00 Explore settings and monitor status from the CMD window
00:13:19 First generation performance, unlimited preset and VRAM use
00:13:57 Second generation speed test after the model is already loaded
00:14:37 Real speed result for generating a full 3.5 minute song
00:14:50 Library tab refresh, date filtering and generation history
00:15:17 Batch folder processing setup and folder logic
00:15:35 Matching lyric/style files and naming rules for batch jobs
00:15:53 Set batch input/output folders and generate by file names
00:16:15 Optional auto-improve tools and LLM lyric instructions file
00:16:45 Built-in enhance lyrics and enhance style options explained
00:17:13 Enhance lyrics example and improved lyric output
00:17:28 Enhance style example and CMD progress tracking
00:17:43 Advanced tab overview: remix controls and reference audio
00:17:58 Advanced lyrics, caption options and generate music button
00:18:22 LoRA folder, refresh button, LoRA scale and preset-linked settings
00:18:54 Repaint feature with source audio plus start and end ranges
00:19:16 Recommended simple workflow for direct song generation
00:19:33 Where to get help, request features and contact support
00:19:53 LoRA training tab and using LLMs for training runs
00:20:27 Dataset preparation and custom presets for saved parameters
00:21:08 Final recap, feature reminder and outro
This video is for anyone who wants a powerful local AI music workflow without relying on paid cloud song generation services. Follow the chapters to install the application, verify your first generation, understand every important setting, and then explore advanced features when you are ready.
For support, questions, feature requests, installation issues or future updates, check the links you add to this description, the pinned comment, Patreon, Discord and the video comments. Thank you for watching and enjoy creating music with ACEStep 1.5 XL Premium.
-
00:00:00 Greetings everyone. Today I am going to introduce you state-of-the-art, open-source,
-
00:00:06 commercially usable, locally runnable music and song generation application,
-
00:00:12 ACEStep 1.5 XL Premium. This application is even better than paid Suno version 5. Yes,
-
00:00:24 you heard it right. There are benchmarks officially published as a research paper,
-
00:00:31 and when we analyze the benchmarks, we see that ACEStep 1.5 XL is better than Suno version 5 at
-
00:00:42 various benchmarks. As you can see, the ACE is scoring best among version 5 and version 4.5.
-
00:00:53 However, there aren't statistics and benchmarks for version 5.5 yet. Even this quality is amazing.
-
00:01:02 So let me show you a quick generation. This is default generation. When you
-
00:01:07 install the application with 1 click on your Windows computer and just hit generate song,
-
00:01:13 this is the quality you are going to get. So we have a demo lyrics and demo style. They will
-
00:01:20 be automatically loaded with the application, and it will automatically pick according to your GPU.
-
00:01:27 You see we have GPU tiers fully automatically, and the default model is ACEStep XL 1.5 Turbo.
-
00:01:35 So let's listen the song and proceed installation setup and all the details that our application
-
00:01:42 has. But before starting, I also want to mention you that we fully support remix and repaint that
-
00:01:51 only paid Suno 5 has. So these are extra features of our application. You don't need to use them
-
00:01:58 if you don't want, but I am just letting you know before we start this amazing tutorial.
-
00:02:14 Woke up with the fire, put the weight on my back. No gold on my wrist, still I'm dressed in facts.
-
00:02:20 Hands in the dirt, put the work on track. Built from the ground up, a life on lack.
-
00:02:25 I don't need the noise, I don't move for praise. I just keep it pure in a world of fakes. Calm in
-
00:02:31 the storm, let the heart stay brave. Letting good seeds let the roots make waves. I came with peace,
-
00:02:37 I came with light. Hands stay clean and my soul stays right. Work all day still I move real kind.
-
00:02:44 Head held low but I aim sky high. No big talk, I don't need that crown. Strong stand tall,
-
00:02:50 keep it humble now. Brick by brick, yeah I built this sound. Quiet on top, let the truth get loud.
-
00:02:57 I was down, I was dust, had to learn my lane. Now I move with grace, never chasing fame. If I eat,
-
00:03:03 we eat, let the whole team gain. Put love in the soil, let it heal that pain. Never flex too hard,
-
00:03:10 let the cold stay deep. Make peace with the grind while the world can't sleep. Heart stay
-
00:03:15 soft but the mind don't creep. What I earn, what I keep, what I sow, I reap. I came with peace,
-
00:03:21 I came with light. Hands stay clean and my soul stays right. Work all day still I move real kind.
-
00:03:28 Head held low but I aim sky high. No big talk, I don't need that crown. Strong stand tall,
-
00:03:34 keep it humble now. Brick by brick, yeah I built this sound. Quiet on top, let the truth get loud.
-
00:03:52 I don't want to shine if my people can't glow. I don't want the wind if it cost my soul.
-
00:03:58 Real ones rise but they still stay low. Deep roots grow where the wild winds blow. I got faith in the
-
00:04:04 good, got strength in the calm. Got scars on my skin but peace in my arms. I don't move in hate, I
-
00:04:10 don't feed that dark. I just light 1 flame, let it spread through hearts. I came with peace, I came
-
00:04:16 with light. Hands stay clean and my soul stays right. Work all day still I move real kind. Head
-
00:04:25 held low but I aim sky high. No big talk, I don't need that crown. Strong stand tall, keep it humble
-
00:04:34 now. Brick by brick, yeah I built this sound. Quiet on top, let the truth get loud. I came with
-
00:04:43 peace, I came with light. Hands stay clean and my soul stays right. Work all day still I move real
-
00:04:52 kind. Head held low but I aim sky high. No big talk, I don't need that crown. Strong stand tall,
-
00:05:01 keep it humble now. Brick by brick, yeah I built this sound. Quiet on top, let the truth get loud.
-
00:05:10 As you have seen, this application is magnificent. This is top tier level quality, fully usable in
-
00:05:20 your commercial products, whatever you need for. So now I will begin with installation and show you
-
00:05:27 all the features we have. This is a super easy to use, extremely well developed application by me,
-
00:05:35 and the installation will be fully 1 click. This is the post where you will download the latest
-
00:05:42 zip file to install. The link will be in the description of the video below, check below and
-
00:05:47 check the pinned comment. Download the latest zip file. The latest zip file version can be
-
00:05:53 lower than the application version. Always check the application version from here, because we
-
00:06:00 are updating from remote server, so it is fine. I recommend you to read this post because this post
-
00:06:07 includes everything you need. Before beginning the installation, make sure that you have followed the
-
00:06:15 Windows requirements 1 time. This is mandatory. You need to have these installed by yourself.
-
00:06:22 Then you can begin the installation. We support RunPod, Massed Compute, and SimplePod as well,
-
00:06:30 and also Linux as well. Those doesn't need requirements tutorial to be followed,
-
00:06:37 they are just ready to use. Move the zip file wherever you want to install. Make sure
-
00:06:42 that you don't have space characters and weird characters in your folder path. So I am going
-
00:06:48 to install into my G drive with test 4 folder. You can install anywhere. Extract files from zip file.
-
00:06:59 And you will get all these files: demo image, ACEStep XL 1.5 SVG file. The installation file
-
00:07:06 is Windows install or update dot bat file. Always use this file to install or update.
-
00:07:13 It will generate a virtual environment and install all the libraries into that folder
-
00:07:20 fully automatically for you. I have pre-compiled most of the super hard to install libraries,
-
00:07:27 and the installation will be also super fast because we are using uv. You can see that we have
-
00:07:32 already generated the virtual environment, so this application installation will not impact anything.
-
00:07:39 Moreover, it will also automatically download the Turbo model as a default. You see we are
-
00:07:47 supporting ACEStep XL 1.5 SFT model, ACEStep XL 1.5 Turbo model, and ACEStep XL 1.5 Base model.
-
00:07:56 Whenever you change these models, it will automatically load best parameters according
-
00:08:02 to your GPU tier. Setting best parameters is really hard for this model, but I have done
-
00:08:09 extensive research and development, so it will be fully automatically set for you. And you can
-
00:08:15 pick the tier according to your GPU, it will load all the best settings with that tier GPU.
-
00:08:23 You can change the vocal language from here. It supports so many languages as you can see. You can
-
00:08:29 change the vocal to male, to female. You can make it instrumental. You can also set song duration.
-
00:08:36 If you don't set it, it will automatically set it according to your lyrics as I have shown you.
-
00:08:42 You can set number of songs to generate. It will generate them as a loop, therefore it won't use
-
00:08:51 bigger VRAM. So you can set as like 10 songs, it will generate 10 different songs with your lyrics.
-
00:09:00 And you can pick the best one among them. The random seed will make each generation different.
-
00:09:07 You can also set custom seed if you want reproducibility. You can set an image to generate
-
00:09:15 a video. So it will generate a song with a video MP4 file. You can set the resolution from here,
-
00:09:23 and it will automatically keep the input image aspect ratio. So you won't need to spend any time
-
00:09:31 to set it accurately. You can click open outputs folder. Every generation will be saved with full
-
00:09:38 metadata, MP3, FLAC files. So whatever you need will be fully saved here. You can later check
-
00:09:45 every generation related file from outputs folder. Every generation will be fully saved automatically
-
00:09:52 for you. So our fresh installation is also completed. You can check all the logs like this,
-
00:10:00 and it also automatically downloaded the necessary models with our special downloader with 16
-
00:10:06 connections and hash verification. Therefore they will be always perfect. Now I will start the new
-
00:10:14 fresh application. To starting the application, click Windows start app dot bat file. It will
-
00:10:21 start the application automatically for you and open the interface. So we got our fresh
-
00:10:27 application. Currently I only have ACEStep XL 1.5 Turbo model. If you want other models,
-
00:10:36 then all you need to do is Windows download all models dot bat file. Run it, it will download the
-
00:10:43 missing models automatically for you. It will also verify the hash values of the previous models and
-
00:10:51 verify that they are all accurately downloaded. So for any reason if your download fails, if your
-
00:10:58 download interrupted, you can double click and run this bat file or the install file to fix your
-
00:11:05 download. You can see the speed, it is amazing. It is 100 megabytes per second. This is my maximum
-
00:11:12 internet speed, so it is using it entirely. To install on Massed Compute and RunPod or SimplePod,
-
00:11:20 open the Massed Compute instructions TXT file. With Massed Compute installation, it
-
00:11:26 will automatically download all of the models, not just Turbo model, because the internet is faster.
-
00:11:32 Just use this link to register. I appreciate that. Then run this, and once the installation is
-
00:11:39 completed, run this. It is so easy. You can follow any of my other Massed Compute tutorials to learn
-
00:11:45 how to use. There is also a tutorial here. So follow the instructions. For RunPod and SimplePod,
-
00:11:52 open this RunPod and SimplePod instructions TXT file. Please use these links to register
-
00:11:59 to SimplePod, and this is for RunPod. On SimplePod you need to use this template,
-
00:12:06 and on RunPod you need to use this template. With RunPod make sure that you filter by CUDA,
-
00:12:13 because we are using CUDA 13, and if you don't filter like this, you won't get newer NVIDIA
-
00:12:19 driver having pod and it will not work. So just read the instructions. It is exactly same as any
-
00:12:28 of my other RunPod or SimplePod tutorial. If you have watched any of them or you can
-
00:12:32 watch these ones to learn. It is so simple to install and use. The usage is totally same on
-
00:12:39 Windows and on RunPod or Massed Compute. The Linux users should follow Massed Compute instructions
-
00:12:46 and Massed Compute install dot sh files. Okay, so this is our fresh installation. As a beginning,
-
00:12:52 I recommend you to generate song and verify that it is working. Once you verified the working,
-
00:13:00 you are ready to go. You can explore the different new settings. You can try different new stuff
-
00:13:07 and see how it is working for your taste. So you can follow the status of what is happening
-
00:13:13 in the very bottom here or on the CMD window. I recommend the CMD window. The generations are
-
00:13:19 ultra fast after the initial generation. It also depends on your selected preset. Currently we are
-
00:13:27 having the unlimited preset. This is for RTX 4090. You can see that it has used about 28 gigabytes.
-
00:13:35 But don't worry, we have GPU tiers for every GPU. The quality may get affected based on
-
00:13:42 the GPU tier. So if you want the highest quality, you can rent RunPod or Massed Compute. But still
-
00:13:48 like 20 gigabyte should also work very well, highest quality as the 24 gigabyte plus models.
-
00:13:57 Okay, the first generation is almost completed. Then I will generate another one to show you
-
00:14:02 the actual speed. Okay, it is generated. Now let's generate another one to see. Okay, I will just
-
00:14:11 hit generate. The generation will immediately start because all the models were already loaded.
-
00:14:17 You see it is generating everything right now. This will take like 1 minute to generate song,
-
00:14:23 maybe even lesser. We are about to see that. Okay, this process takes like 30 seconds. It may get
-
00:14:30 faster, completed as well. Yes, after the first 30 seconds, it is almost done.
-
00:14:37 Yes. So it is taking around 40 seconds to generate 3.5 minutes long song. It is amazing speed. It
-
00:14:46 is like real time 3x, 4x speed. So when you go to library, click refresh, and you can filter by date
-
00:14:54 and see every one of your generated songs. This is super easy to use to see your older generations,
-
00:15:01 filter them by date, and see all of their data like lyrics, full metadata, whatever you have used
-
00:15:09 to generate that song. We also have batch folder processing. Batch folder processing logic is
-
00:15:17 so simple. So generate a folder like this, put your lyrics like this: song 1 TXT file,
-
00:15:28 and copy paste your lyric, whatever the lyric you have, like this one. Then for style,
-
00:15:35 generate a file named like this: song 1 style, and put your styles like this. They can be any name
-
00:15:43 like GG and then you need to have GG style. So it will read every file you given here,
-
00:15:53 and it will generate into the outputs folder with that name. So how you are going to give that? This
-
00:16:01 will be input folder, and wherever you want to generate like outputs, and it will generate songs
-
00:16:08 with the file names. It will use the lyrics and the style files. You can also auto improve lyrics
-
00:16:15 and auto improve style. However, I don't recommend that. I think it is not working very well.
-
00:16:21 So what else you can do? We have a file, a special file in the zip file. It is called as
-
00:16:30 ACEStep lyrics generation instructions for LLMs. Use your favorite LLM like ChatGPT, Gemini,
-
00:16:37 Claude, whatever. Provide this file and provide your lyrics to make it better, better formatted.
-
00:16:45 I recommend this, but still you can use this enhanced style, enhanced lyrics. This is using the
-
00:16:51 models that we already downloaded with the application. This is made by the ACEStep
-
00:16:56 developers. I am giving you the option. You can test each of them and like or not. For example,
-
00:17:04 let's try just these lyrics and see how it enhance lyrics. Enhance style will enhance this style.
-
00:17:13 Enhance lyrics will improve lyrics. Okay, so I just given this and it has enhanced it like this,
-
00:17:21 as you can see. So you can do whatever you want. Let's try just rap for enhanced style and see.
-
00:17:28 It was also pretty fast. It takes a few seconds to enhance. You can always follow what is happening
-
00:17:36 on the CMD window, so it has enhanced the rap into this one. Okay, the advanced tab has so many
-
00:17:43 features. For example, you can use remix. Read everything here. You see the source composition,
-
00:17:50 retake, edit, remix strength, no FSQ, reference audio, music caption and lyrics. It will use
-
00:17:58 the caption and lyrics from here. You can use enhance here too, instrumental here.
-
00:18:04 And optional parameters, just read them, and if you want to apply them, apply them.
-
00:18:11 You can also pick these options and then hit the generate music in the advanced tab to generate.
-
00:18:17 The generated music will also appear here. You can also see the other parameters. We
-
00:18:22 all automatically set these parameters for you. You can also use LoRA. We have LoRA folder. Just
-
00:18:30 put the LoRAs into this folder, LoRAs folder, and click refresh. The LoRAs folder will be inside the
-
00:18:38 installation, inside LoRAs. So here, just put your LoRAs, and there is LoRA scale and other stuff.
-
00:18:45 So all of these are automatically set with our model selection and the VRAM preset. Therefore,
-
00:18:54 it is up to you whichever you want to use. Moreover, you can also use repaint option in the
-
00:19:00 advanced. You provide source audio, pay attention to every detail as I said. Read everything. The
-
00:19:08 repainting has repaint start and repaint end. This application has so many features as I have said.
-
00:19:16 You can explore all of them. All of them is tested and working. But if you just want to
-
00:19:21 generate amazing songs and use them right away, this is the tab that you need. This is a fully
-
00:19:27 developed application by me. You will not find this quality application anywhere else.
-
00:19:33 And if you have ever any problem or request, you can message me from Patreon,
-
00:19:39 from leaving a comment to this video. You can contact me from anywhere, from our Discord server.
-
00:19:46 I am very active on Discord server. And you can let me know whatever you need, help or a new
-
00:19:53 feature or a request. This application also has LoRA training. It is working. You can use your
-
00:20:00 favorite LLM like Codex, like Gemini or Claude to get the training run. It also has all the
-
00:20:09 information that you need. I don't have time to research and show you how to train at the moment,
-
00:20:16 but it is working as a LoRA training. You can train your LoRA and possibly obtain a consistent
-
00:20:27 audio and music generation. So you can also train your LoRA to improve consistency. The dataset
-
00:20:35 tab is for preparing your LoRA training dataset. And we have custom preset. What is the purpose of
-
00:20:42 custom preset? Let's say you want to change some of the parameters and save them, like disable
-
00:20:49 this, this or duration auto and send and set like this. You can save everything with custom preset
-
00:20:57 like GG and save preset, and it will save the preset. And when you restart your application,
-
00:21:02 it will automatically load that preset next time. You can also delete all the presets and use the
-
00:21:08 default presets. So this is all. All the features are like this. Explore the application and let me
-
00:21:16 know. Thank you so much for watching. Hopefully see you in another amazing tutorial because I am
-
00:21:21 working on multiple applications and each one is taking to perfect them. Thank you, see you later.
